CN104361366B - License plate recognition method and license plate recognition device - Google Patents
License plate recognition method and license plate recognition device Download PDFInfo
- Publication number
- CN104361366B CN104361366B CN201410742690.8A CN201410742690A CN104361366B CN 104361366 B CN104361366 B CN 104361366B CN 201410742690 A CN201410742690 A CN 201410742690A CN 104361366 B CN104361366 B CN 104361366B
- Authority
- CN
- China
- Prior art keywords
- license plate
- classifier
- training set
- scene
- negative samples
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Traffic Control Systems (AREA)
- Character Discrimination (AREA)
- Image Analysis (AREA)
Abstract
Description
技术领域technical field
本申请涉及图像识别技术领域,特别是涉及一种车牌识别方法及车牌识别设备。The present application relates to the technical field of image recognition, in particular to a license plate recognition method and a license plate recognition device.
背景技术Background technique
为了保证良好的交通秩序或社会治安,基于车牌识别技术的车牌识别设备等产品被广泛应用于交叉路口、物业小区、商业楼宇及政府机构等位置。车牌识别技术一般分为车牌检测、字符分割识别及车牌投票这三大步骤。车牌检测,即从视频中检测车牌并确定其在每一帧图像中的位置,是车牌识别技术中比较关键和耗时的一步。In order to ensure good traffic order or social security, products such as license plate recognition equipment based on license plate recognition technology are widely used in intersections, property communities, commercial buildings and government agencies. License plate recognition technology is generally divided into three steps: license plate detection, character segmentation and recognition, and license plate voting. License plate detection, that is, detecting the license plate from the video and determining its position in each frame of the image, is a critical and time-consuming step in license plate recognition technology.
目前,车牌识别设备通常都是在实验室进行训练。通常情况下,车辆正面只有一个车牌,并且在路口、岗亭或卡口,车辆都是依次通行的,通过使用摄像机在各个时段和天气条件下在各个位置拍摄并储存大量视频,在这些视频的图像中,人工截取出每辆车的车牌图像作为正样本,人工截取出其中的非车牌图像作为负样本,然后通过正、负样本训练得到车牌识别设备。At present, license plate recognition equipment is usually trained in the laboratory. Usually, there is only one license plate on the front of the vehicle, and at intersections, sentry boxes or bayonets, vehicles pass in turn. By using cameras to shoot and store a large number of videos at various locations under various time periods and weather conditions, the images in these videos In this method, the license plate image of each vehicle is manually intercepted as a positive sample, and the non-license plate image is manually intercepted as a negative sample, and then the license plate recognition device is obtained through positive and negative sample training.
然而,在实际应用中,现场环境和应用情况千变万化,车牌识别设备在实验室训练时由于负样本的数量和种类有限,无法兼顾所有的应用场景和工作时段,因而车牌识别设备在现场使用的时候会存在一定的误检,即把一些非车牌图像错误地识别成车牌。误检包括静态的误检和动态的误检。静态的误检是指现场的场景中一些背景,例如路边的栅栏,广告牌上的电话号码等,这些背景图案与实际的车牌很相似,车牌识别设备很容易把这些背景错误地识别成车牌,比如路边的栅栏与“111111”很相似,容易被识别成“省份L11111”,另外一些字符如“1”、“L”、“T”等也容易被错误地识别成“H”或“Y”。动态的误检是指移动的物体,例如汽车引擎的通风口、车身上的广告等,这些物体偶尔会出现,由于比较类似车牌,也会造成一定的误检,如果车牌识别设备应用在快递公司出入口,公交公司停车场等场所时,由于大量的车身上都具有广告和电话号码等,就容易出现大量误检,导致车牌识别出错。However, in practical applications, the on-site environment and application conditions are ever-changing. Due to the limited number and types of negative samples in the laboratory training, the license plate recognition equipment cannot take into account all application scenarios and working hours. Therefore, when the license plate recognition equipment is used in the field There will be some false detection, that is, some non-license plate images are mistakenly recognized as license plates. False detection includes static false detection and dynamic false detection. Static false detection refers to some backgrounds in the on-site scene, such as roadside fences, phone numbers on billboards, etc. These background patterns are very similar to actual license plates, and license plate recognition equipment can easily mistake these backgrounds as license plates , For example, the fence on the side of the road is very similar to "111111", which is easily recognized as "Province L11111", and other characters such as "1", "L", "T", etc. are also easily recognized as "H" or " Y". Dynamic false detection refers to moving objects, such as car engine vents, advertisements on the car body, etc. These objects occasionally appear, and because they are more similar to license plates, they will also cause certain false detections. If the license plate recognition equipment is used in express companies At entrances and exits, bus company parking lots and other places, because a large number of car bodies have advertisements and phone numbers, etc., it is easy to have a large number of false detections, resulting in errors in license plate recognition.
发明内容Contents of the invention
有鉴于此,本申请提供一种车牌识别方法及车牌识别设备,以实现对车牌的精确识别,减少车牌识别设备的误检率。In view of this, the present application provides a license plate recognition method and a license plate recognition device, so as to realize accurate recognition of the license plate and reduce the false detection rate of the license plate recognition device.
为了实现上述目的,本申请实施例提供的技术方案如下:In order to achieve the above objectives, the technical solutions provided in the embodiments of the present application are as follows:
一种车牌识别方法,应用于布置在现场的车牌识别设备中,所述车牌识别设备中包括根据原始训练集训练的原始车牌分类器,所述车牌识别方法包括:A license plate recognition method, which is applied to a license plate recognition device arranged at the scene, wherein the license plate recognition device includes an original license plate classifier trained according to an original training set, and the license plate recognition method includes:
根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本;所述静态负样本为所述现场的无车场景中被所述原始车牌分类器误检为车牌区域的背景图像;According to the car-free scene of the scene detected by the original license plate classifier, the static negative samples of the scene are collected; ;
将所述静态负样本添加到所述车牌识别设备的所述原始训练集中,得到第一训练集,根据所述第一训练集训练第一车牌分类器;Adding the static negative sample to the original training set of the license plate recognition device to obtain a first training set, and training a first license plate classifier according to the first training set;
根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本,将所述动态负样本添加到所述第一训练集中,得到第二训练集,根据所述第二训练集训练第二车牌分类器,并根据所述第二车牌分类器进行车牌识别;所述动态负样本为所述现场的有车场景中被所述第一车牌分类器误检为车牌区域的运动图像。According to the first license plate classifier to detect the car scene on the scene, collect the dynamic negative samples on the scene, add the dynamic negative samples to the first training set to obtain the second training set, according to the second training set Train the second license plate classifier, and carry out license plate recognition according to the second license plate classifier; the dynamic negative sample is a moving image of the license plate area misdetected by the first license plate classifier in the scene with a car in the scene .
优选地,所述根据所述第一训练集训练第一车牌分类器,包括:Preferably, the training of the first license plate classifier according to the first training set includes:
使用Haar特征对所述第一训练集里的每一个正样本进行表征,形成正样本Haar特征向量;Using the Haar feature to characterize each positive sample in the first training set to form a positive sample Haar feature vector;
使用Haar特征对所述第一训练集里的每一个负样本进行表征,形成负样本Haar特征向量;Each negative sample in the first training set is characterized using a Haar feature to form a negative sample Haar feature vector;
利用Adaboost算法对所述正样本Haar特征向量和所述负样本Haar特征向量进行训练,获得第一车牌分类器。The Adaboost algorithm is used to train the positive sample Haar feature vector and the negative sample Haar feature vector to obtain a first license plate classifier.
优选地,所述根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本,包括:Preferably, the scene of having a car on the scene is detected according to the first license plate classifier, and the dynamic negative samples of the scene are collected, including:
获取所述车牌识别设备检测现场的有车场景得到的视频图像,提取所述视频图像中的疑似车牌区域,根据所述第一车牌分类器检测所述疑似车牌区域中的车牌区域;Obtaining the video image obtained by the vehicle scene detected by the license plate recognition device, extracting the suspected license plate area in the video image, and detecting the license plate area in the suspected license plate area according to the first license plate classifier;
从所述车牌区域中分割出多个字符,根据支持向量机SVM训练的车牌字符识别模型对多个所述字符进行识别,并判断每个所述字符的识别置信度;Segment a plurality of characters from the license plate area, identify a plurality of the characters according to the license plate character recognition model trained by the support vector machine SVM, and judge the recognition confidence of each of the characters;
根据每个所述字符的识别置信度判断所述车牌区域是否有效,如果无效,将所述车牌区域作为所述现场的动态负样本。Whether the license plate area is valid is judged according to the recognition confidence of each character, and if invalid, the license plate area is used as a dynamic negative sample of the scene.
优选地,所述根据每个所述字符的识别置信度判断所述车牌区域是否有效,包括:Preferably, the judging whether the license plate area is valid according to the recognition confidence of each character includes:
判断所述字符的个数是否为7个;Determine whether the number of characters is 7;
如果是,判断每个所述字符的识别置信度是否大于或等于第一阈值;If so, judging whether the recognition confidence of each character is greater than or equal to a first threshold;
如果是,判断7个所述字符的识别置信度的和是否大于或等于第二阈值;If yes, judging whether the sum of the recognition confidences of the 7 characters is greater than or equal to the second threshold;
如果是,则判定所述车牌区域有效,否则无效。If so, it is determined that the license plate area is valid, otherwise it is invalid.
优选地,所述将所述动态负样本添加到所述第一训练集中,包括:Preferably, the adding the dynamic negative samples to the first training set includes:
按照预设的时间间隔统计所述第一车牌分类器检测现场的有车场景的误检率,并判断所述误检率是否大于第三阈值;Counting the false detection rate of the car scene detected by the first license plate classifier according to the preset time interval, and judging whether the false detection rate is greater than the third threshold;
如果是,判断收集到的所述动态负样本的个数是否大于或等于第四阈值;If yes, determine whether the number of the collected dynamic negative samples is greater than or equal to the fourth threshold;
如果是,将所述动态负样本添加到所述第一训练集中。If yes, add the dynamic negative samples to the first training set.
优选地,在所述根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本后,还包括:Preferably, after detecting the car-free scene of the scene according to the original license plate classifier, and collecting the static negative samples of the scene, it also includes:
识别所述静态负样本中的简单静态负样本和困难静态负样本,并将所述简单静态负样本添加到所述车牌识别设备的所述原始训练集中;identifying easy static negative samples and difficult static negative samples in the static negative samples, and adding the easy static negative samples to the original training set of the license plate recognition device;
判断所述困难静态负样本所在的图像区域是否满足屏蔽要求;Judging whether the image region where the difficult static negative sample is located meets the masking requirement;
如果满足,屏蔽所述困难静态负样本所在的图像区域,如果不满足,根据所述困难静态负样本训练第三车牌分类器,所述第三车牌分类器用于判断所述第二车牌分类器识别出的车牌是否为有效车牌。If it is satisfied, shield the image area where the difficult static negative sample is located, if not, train a third license plate classifier according to the difficult static negative sample, and the third license plate classifier is used to judge the identification of the second license plate classifier Whether the issued license plate is a valid license plate.
优选地,在所述根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本后,还包括:Preferably, after detecting the car-free scene of the scene according to the original license plate classifier, and collecting the static negative samples of the scene, it also includes:
识别所述静态负样本中的简单静态负样本和困难静态负样本,并将所述简单静态负样本添加到所述车牌识别设备的所述原始训练集中;identifying easy static negative samples and difficult static negative samples in the static negative samples, and adding the easy static negative samples to the original training set of the license plate recognition device;
判断所述困难静态负样本所在的图像区域是否满足屏蔽要求;Judging whether the image region where the difficult static negative sample is located meets the masking requirement;
如果满足,屏蔽所述困难静态负样本所在的图像区域,如果不满足,在预设的时间段内判断所述困难静态负样本所在的图像区域是否发生移动,如果不发生移动,则将所述困难静态负样本添加到所述车牌识别设备的所述原始训练集中。If it is satisfied, shield the image area where the difficult static negative sample is located. If it is not satisfied, judge whether the image area where the difficult static negative sample is located moves within a preset period of time. If no movement occurs, the Difficult static negative samples are added to the original training set of the license plate recognition device.
本申请还提供一种车牌识别设备,用于布置在现场进行车牌识别,所述车牌识别设备中包括根据原始训练集训练的原始车牌分类器,所述车牌识别设备还包括:The present application also provides a license plate recognition device, which is used to arrange on-site license plate recognition. The license plate recognition device includes an original license plate classifier trained according to the original training set, and the license plate recognition device also includes:
静态负样本收集模块,用于根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本;所述静态负样本为所述现场的无车场景中被所述原始车牌分类器误检为车牌区域的背景图像;The static negative sample collection module is used to detect the car-free scene of the scene according to the original license plate classifier, and collect the static negative sample of the scene; It is falsely detected as the background image of the license plate area;
第一车牌分类器模块,用于将所述静态负样本添加到所述车牌识别设备的所述原始训练集中,得到第一训练集,根据所述第一训练集训练第一车牌分类器;The first license plate classifier module is used to add the static negative sample to the original training set of the license plate recognition device to obtain a first training set, and train the first license plate classifier according to the first training set;
第二车牌分类器模块,用于根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本,将所述动态负样本添加到所述第一训练集中,得到第二训练集,根据所述第二训练集训练第二车牌分类器,并根据所述第二车牌分类器进行车牌识别;所述动态负样本为所述现场的有车场景中被所述第一车牌分类器误检为车牌区域的运动图像。The second license plate classifier module is used to detect the on-site car scene according to the first license plate classifier, collect dynamic negative samples on the scene, add the dynamic negative samples to the first training set, and obtain the second training set, train a second license plate classifier according to the second training set, and perform license plate recognition according to the second license plate classifier; The detector misdetects it as a moving image in the license plate area.
优选地,所述第一车牌分类器模块,包括:Preferably, the first license plate classifier module includes:
正样本特征向量单元,用于使用Haar特征对所述第一训练集里的每一个正样本进行表征,形成正样本Haar特征向量;A positive sample feature vector unit, configured to characterize each positive sample in the first training set using Haar features to form a positive sample Haar feature vector;
负样本特征向量单元,用于使用Haar特征对所述第一训练集里的每一个负样本进行表征,形成负样本Haar特征向量;Negative sample feature vector unit, for using Haar feature to represent each negative sample in the first training set to form a negative sample Haar feature vector;
训练单元,用于利用Adaboost算法对所述正样本Haar特征向量和所述负样本Haar特征向量进行训练,获得第一车牌分类器。The training unit is configured to use the Adaboost algorithm to train the positive sample Haar feature vector and the negative sample Haar feature vector to obtain a first license plate classifier.
优选地,所述第二车牌分类器模块,包括:Preferably, the second license plate classifier module includes:
检测单元,用于获取所述车牌识别设备检测现场的有车场景得到的视频图像,提取所述视频图像中的疑似车牌区域,根据所述第一车牌分类器检测所述疑似车牌区域中的车牌区域;A detection unit, configured to acquire a video image obtained by the vehicle scene detected by the license plate recognition device, extract the suspected license plate area in the video image, and detect the license plate in the suspected license plate area according to the first license plate classifier area;
识别单元,用于从所述车牌区域中分割出多个字符,根据支持向量机SVM训练的车牌字符识别模型对多个所述字符进行识别,并判断每个所述字符的识别置信度;A recognition unit, configured to segment a plurality of characters from the license plate area, recognize a plurality of the characters according to the license plate character recognition model trained by the support vector machine SVM, and judge the recognition confidence of each of the characters;
动态负样本单元,用于根据每个所述字符的识别置信度判断所述车牌区域是否有效,如果无效,将所述车牌区域作为所述现场的动态负样本。The dynamic negative sample unit is used to judge whether the license plate area is valid according to the recognition confidence of each character, and if invalid, use the license plate area as a dynamic negative sample of the scene.
优选地,所述动态负样本单元,包括:Preferably, the dynamic negative sample unit includes:
第一判断子单元,用于判断所述字符的个数是否为7个;The first judging subunit is used to judge whether the number of the characters is 7;
第二判断子单元,用于如果是,判断每个所述字符的识别置信度是否大于或等于第一阈值;The second judging subunit is configured to, if yes, judging whether the recognition confidence of each character is greater than or equal to the first threshold;
第三判断子单元,用于如果是,判断7个所述字符的识别置信度的和是否大于或等于第二阈值;The third judging subunit is used to judge whether the sum of the recognition confidences of the 7 characters is greater than or equal to the second threshold;
判定子单元,用于如果是,则判定所述车牌区域有效,否则无效。A judging subunit, configured to judge that the license plate area is valid if yes, otherwise it is invalid.
优选地,所述第二车牌分类器模块,包括:Preferably, the second license plate classifier module includes:
统计单元,用于按照预设的时间间隔统计所述第一车牌分类器检测现场的有车场景的误检率,并判断所述误检率是否大于第三阈值;A statistical unit, configured to count the false detection rate of the car scene detected by the first license plate classifier according to a preset time interval, and determine whether the false detection rate is greater than a third threshold;
判断单元,用于如果是,判断收集到的所述动态负样本的个数是否大于或等于第四阈值;a judging unit, configured to, if yes, judge whether the number of the collected dynamic negative samples is greater than or equal to a fourth threshold;
添加单元,用于如果是,将所述动态负样本添加到所述第一训练集中。The adding unit is configured to, if yes, add the dynamic negative samples to the first training set.
优选地,所述静态负样本收集模块,包括:Preferably, the static negative sample collection module includes:
识别单元,用于识别所述静态负样本中的简单静态负样本和困难静态负样本,并将所述简单静态负样本添加到所述车牌识别设备的所述原始训练集中;A recognition unit, configured to identify simple static negative samples and difficult static negative samples in the static negative samples, and add the simple static negative samples to the original training set of the license plate recognition device;
判断单元,用于判断所述困难静态负样本所在的图像区域是否满足屏蔽要求;A judging unit, configured to judge whether the image region where the difficult static negative sample is located satisfies the masking requirement;
屏蔽单元,用于如果满足,屏蔽所述困难静态负样本所在的图像区域,如果不满足,根据所述困难静态负样本训练第三车牌分类器,所述第三车牌分类器用于判断所述第二车牌分类器识别出的车牌是否为有效车牌。The shielding unit is used to shield the image area where the difficult static negative sample is located if it is satisfied, and if it is not satisfied, train a third license plate classifier according to the difficult static negative sample, and the third license plate classifier is used to judge the first Second, whether the license plate recognized by the license plate classifier is a valid license plate.
优选地,所述静态负样本收集模块,包括:Preferably, the static negative sample collection module includes:
识别单元,用于识别所述静态负样本中的简单静态负样本和困难静态负样本,并将所述简单静态负样本添加到所述车牌识别设备的所述原始训练集中;A recognition unit, configured to identify simple static negative samples and difficult static negative samples in the static negative samples, and add the simple static negative samples to the original training set of the license plate recognition device;
判断单元,用于判断所述困难静态负样本所在的图像区域是否满足屏蔽要求;A judging unit, configured to judge whether the image region where the difficult static negative sample is located satisfies the masking requirement;
屏蔽单元,用于如果满足,屏蔽所述困难静态负样本所在的图像区域,如果不满足,在预设的时间段内判断所述困难静态负样本所在的图像区域是否发生移动,如果不发生移动,则将所述困难静态负样本添加到所述车牌识别设备的所述原始训练集中。The shielding unit is used to shield the image area where the difficult static negative sample is located if it is satisfied, and if it is not satisfied, judge whether the image area where the difficult static negative sample is located moves within a preset time period, and if it does not move , then add the difficult static negative samples to the original training set of the license plate recognition device.
由以上本申请提供的技术方案,车牌识别方法应用于布置在现场的车牌识别设备中,所述车牌识别设备中包括根据原始训练集训练的原始车牌分类器,所述车牌识别方法根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本;所述静态负样本为所述现场的无车场景中被所述原始车牌分类器误检为车牌区域的背景图像;将所述静态负样本添加到所述车牌识别设备的所述原始训练集中,得到第一训练集,根据所述第一训练集训练第一车牌分类器;根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本,将所述动态负样本添加到所述第一训练集中,得到第二训练集,根据所述第二训练集训练第二车牌分类器,并根据所述第二车牌分类器进行车牌识别;所述动态负样本为所述现场的有车场景中被所述第一车牌分类器误检为车牌区域的运动图像。这样,先收集现场的静态负样本并添加到原始训练集,得到第一训练集,利用第一训练集训练第一车牌分类器去除静态负样本,然后再使用第一车牌分类器收集动态负样本并添加到第一训练集,得到第二训练集,利用第二训练集训练第二车牌分类器就可以去除收集到的静态负样本和动态负样本,此时第二训练集中已经包含了现场可能出现的绝大部分负样本,这些负样本能够被第二车牌分类器拒绝,从而达到对车牌的精确识别,减少车牌识别设备的误检率。Based on the above technical solution provided by the present application, the license plate recognition method is applied to the license plate recognition equipment arranged on site, and the license plate recognition equipment includes an original license plate classifier trained according to the original training set, and the license plate recognition method is based on the original The license plate classifier detects the car-free scene of the scene, and collects the static negative samples of the scene; the static negative sample is the background image of the license plate area misdetected by the original license plate classifier in the car-free scene of the scene; Static negative samples are added to the original training set of the license plate recognition device to obtain a first training set, and a first license plate classifier is trained according to the first training set; Scenario, collect the dynamic negative samples of the scene, add the dynamic negative samples to the first training set to obtain the second training set, train the second license plate classifier according to the second training set, and according to the second training set The license plate classifier performs license plate recognition; the dynamic negative sample is a moving image in the on-site vehicle scene that is misdetected as a license plate area by the first license plate classifier. In this way, first collect the static negative samples on the scene and add them to the original training set to obtain the first training set, use the first training set to train the first license plate classifier to remove static negative samples, and then use the first license plate classifier to collect dynamic negative samples And add it to the first training set to get the second training set, and use the second training set to train the second license plate classifier to remove the collected static negative samples and dynamic negative samples. At this time, the second training set already contains the on-site possible Most of the negative samples that appear, these negative samples can be rejected by the second license plate classifier, so as to achieve accurate recognition of the license plate and reduce the false detection rate of the license plate recognition device.
附图说明Description of drawings
为了更清楚地说明本申请实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请中记载的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments described in this application. Those skilled in the art can also obtain other drawings based on these drawings without creative work.
图1为本申请提供的一种车牌识别方法的流程示意图;Fig. 1 is a schematic flow chart of a license plate recognition method provided by the present application;
图2为本申请提供的另一种车牌识别方法的流程示意图;Fig. 2 is a schematic flow chart of another license plate recognition method provided by the present application;
图3为本申请提供的又一种车牌识别方法的流程示意图;Fig. 3 is a schematic flow chart of another license plate recognition method provided by the present application;
图4为本申请提供的一种车牌识别设备的结构示意图。FIG. 4 is a schematic structural diagram of a license plate recognition device provided by the present application.
具体实施方式Detailed ways
为了使本技术领域的人员更好地理解本申请中的技术方案,下面将结合附图,对本申请的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都应当属于本申请保护的范围。In order to enable those skilled in the art to better understand the technical solutions of the present application, the technical solutions of the present application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application. , but not all examples. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.
下面结合附图,对本申请的实施方案进行详细描述。Embodiments of the present application will be described in detail below in conjunction with the accompanying drawings.
图1为本申请提供的一种车牌识别方法的流程示意图。FIG. 1 is a schematic flowchart of a license plate recognition method provided by the present application.
请参照图1所示,本申请实施例提供一种车牌识别方法,应用于布置在现场的车牌识别设备中,所述车牌识别设备中包括根据原始训练集训练的原始车牌分类器,所述车牌识别方法包括:Please refer to Fig. 1, the embodiment of the present application provides a license plate recognition method, which is applied to the license plate recognition equipment arranged on the scene, the license plate recognition equipment includes the original license plate classifier trained according to the original training set, the license plate Identification methods include:
S100:根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本;S100: According to the original license plate classifier, detect the car-free scene on the scene, and collect the static negative samples on the scene;
所述静态负样本为所述现场的无车场景中被所述原始车牌分类器误检为车牌区域的背景图像;The static negative sample is the background image of the license plate area misdetected by the original license plate classifier in the car-free scene of the scene;
在本申请实施例中,在现场安装车牌识别设备后,首先使用原始车牌分类器检测现场无车场景,收集导致原始车牌分类器误检的图像,作为错分负样本,即静态负样本。In the embodiment of the present application, after the license plate recognition equipment is installed on site, the original license plate classifier is first used to detect the car-free scene on the scene, and the images that cause the original license plate classifier to misdetect are collected as misclassified negative samples, that is, static negative samples.
车牌识别设备安装在现场的时候,要把实验室训练车牌分类器的原始正样本和原始负样本一起保存在现场的计算机内。开启车牌识别器后,在确保现场无车的情况下按照白昼、黑夜、阴天、晴天、强光、弱光等不同环境对现场进行视频流车牌检测,如果能检测到车牌,就表示现场场景中的某些背景被识别成车牌了,此时保留这些错误的样本,即现场的静态负样本。When the license plate recognition equipment is installed on site, the original positive samples and original negative samples of the license plate classifier trained in the laboratory should be saved in the computer on site together. After turning on the license plate recognizer, under the condition of ensuring that there is no car on the scene, the video stream license plate detection is performed on the scene according to different environments such as day, night, cloudy, sunny, strong light, and weak light. If the license plate can be detected, it means the scene of the scene Some of the background in is recognized as a license plate, and these wrong samples are kept at this time, that is, the static negative samples of the scene.
S200:将所述静态负样本添加到所述车牌识别设备的所述原始训练集中,得到第一训练集,根据所述第一训练集训练第一车牌分类器;S200: Add the static negative sample to the original training set of the license plate recognition device to obtain a first training set, and train a first license plate classifier according to the first training set;
在本申请实施例中,将现场的静态负样本添加到原始训练集,得到第一训练集,根据新的训练集即第一训练集训练新的车牌分类器即第一车牌分类器,从而用新的第一车牌分类器替代旧的原始车牌分类器进行车牌检测,就可以不受静态负样本的影响,实现对静态负样本的正确识别。In the embodiment of the present application, the static negative samples of the scene are added to the original training set to obtain the first training set, and a new license plate classifier, namely the first license plate classifier, is trained according to the new training set, namely the first training set, so as to use The new first license plate classifier replaces the old original license plate classifier for license plate detection, which can not be affected by static negative samples and realize the correct identification of static negative samples.
S300:根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本,将所述动态负样本添加到所述第一训练集中,得到第二训练集,根据所述第二训练集训练第二车牌分类器,并根据所述第二车牌分类器进行车牌识别。S300: According to the first license plate classifier to detect the on-site car scene, collect the dynamic negative samples on the scene, add the dynamic negative samples to the first training set to obtain the second training set, and according to the second The training set trains a second license plate classifier, and performs license plate recognition according to the second license plate classifier.
所述动态负样本为所述现场的有车场景中被所述第一车牌分类器误检为车牌区域的运动图像。The dynamic negative samples are motion images in the on-site vehicle scene that are misdetected as the license plate area by the first license plate classifier.
在本申请实施例中,用新的第一车牌分类器替代旧的原始车牌分类器后,在第一车牌分类器的使用过程中收集现场的动态负样本,并把收集的动态负样本添加到第一训练集中,得到第二训练集,重新根据第二训练集训练第二车牌分类器,并根据第二车牌分类器进行车牌识别,就可以不受现场的静态负样本和已经收集到的动态负样本的影响,实现对静态负样本和收集到的动态负样本的正确识别,减少误检率。In the embodiment of this application, after replacing the old original license plate classifier with a new first license plate classifier, the dynamic negative samples on the scene are collected during the use of the first license plate classifier, and the collected dynamic negative samples are added to In the first training set, get the second training set, retrain the second license plate classifier according to the second training set, and perform license plate recognition based on the second license plate classifier, so that it is not affected by the static negative samples on the scene and the collected dynamic The impact of negative samples can realize the correct identification of static negative samples and collected dynamic negative samples, and reduce the false positive rate.
进一步的,在第二车牌分类器的使用过程中,本申请还可以继续收集动态负样本,并将收集的动态负样本继续添加进第二训练集中得到第三训练集,并且根据第三训练集训练第三车牌分类器,根据第三车牌分类器进行车牌识别,依次类推。该方法无需人工干预,经过多次迭代后能有效减少车牌识别设备的误检,后续的迭代方法与步骤300类似,属于同一原理的重复运用,此处不再赘述。Further, during the use of the second license plate classifier, the present application can also continue to collect dynamic negative samples, and continue to add the collected dynamic negative samples to the second training set to obtain the third training set, and according to the third training set Train the third license plate classifier, perform license plate recognition according to the third license plate classifier, and so on. This method does not require manual intervention, and can effectively reduce the false detection of the license plate recognition device after multiple iterations. The subsequent iteration method is similar to step 300, which belongs to the repeated application of the same principle, and will not be repeated here.
由以上本申请提供的技术方案,车牌识别方法应用于布置在现场的车牌识别设备中,所述车牌识别设备中包括根据原始训练集训练的原始车牌分类器,所述车牌识别方法根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本;所述静态负样本为所述现场的无车场景中被所述原始车牌分类器误检为车牌区域的背景图像;将所述静态负样本添加到所述车牌识别设备的所述原始训练集中,得到第一训练集,根据所述第一训练集训练第一车牌分类器;根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本,将所述动态负样本添加到所述第一训练集中,得到第二训练集,根据所述第二训练集训练第二车牌分类器,并根据所述第二车牌分类器进行车牌识别;所述动态负样本为所述现场的有车场景中被所述第一车牌分类器误检为车牌区域的运动图像。这样,先收集现场的静态负样本并添加到原始训练集,得到第一训练集,利用第一训练集训练第一车牌分类器去除静态负样本,然后再使用第一车牌分类器收集动态负样本并添加到第一训练集,得到第二训练集,利用第二训练集训练第二车牌分类器就可以去除收集到的静态负样本和动态负样本,此时第二训练集中已经包含了现场可能出现的绝大部分负样本,这些负样本能够被第二车牌分类器拒绝,从而达到对车牌的精确识别,减少车牌识别设备的误检率。Based on the above technical solution provided by the present application, the license plate recognition method is applied to the license plate recognition equipment arranged on site, and the license plate recognition equipment includes an original license plate classifier trained according to the original training set, and the license plate recognition method is based on the original The license plate classifier detects the car-free scene of the scene, and collects the static negative samples of the scene; the static negative sample is the background image of the license plate area misdetected by the original license plate classifier in the car-free scene of the scene; Static negative samples are added to the original training set of the license plate recognition device to obtain a first training set, and a first license plate classifier is trained according to the first training set; Scenario, collect the dynamic negative samples of the scene, add the dynamic negative samples to the first training set to obtain the second training set, train the second license plate classifier according to the second training set, and according to the second training set The license plate classifier performs license plate recognition; the dynamic negative sample is a moving image in the on-site vehicle scene that is misdetected as a license plate area by the first license plate classifier. In this way, first collect the static negative samples on the scene and add them to the original training set to obtain the first training set, use the first training set to train the first license plate classifier to remove static negative samples, and then use the first license plate classifier to collect dynamic negative samples And add it to the first training set to get the second training set, and use the second training set to train the second license plate classifier to remove the collected static negative samples and dynamic negative samples. At this time, the second training set already contains the on-site possible Most of the negative samples that appear, these negative samples can be rejected by the second license plate classifier, so as to achieve accurate recognition of the license plate and reduce the false detection rate of the license plate recognition device.
上述实施例提供了一种车牌识别方法,其中,根据所述第一训练集训练第一车牌分类器的方法,本实施例将结合附图进行说明:The above-mentioned embodiment provides a kind of license plate recognition method, wherein, according to the method for training the first license plate classifier according to the first training set, this embodiment will be described in conjunction with the accompanying drawings:
图2为本申请提供的另一种车牌识别方法的流程示意图。FIG. 2 is a schematic flowchart of another license plate recognition method provided by the present application.
请参照图2所示,本申请实施例提供的方法,包括:Please refer to Figure 2, the method provided by the embodiment of the present application includes:
S201:使用Haar特征对所述第一训练集里的每一个正样本进行表征,形成正样本Haar特征向量;S201: Using Haar features to characterize each positive sample in the first training set to form a positive sample Haar feature vector;
S202:使用Haar特征对所述第一训练集里的每一个负样本进行表征,形成负样本Haar特征向量;S202: Using Haar features to characterize each negative sample in the first training set to form a negative sample Haar feature vector;
S203:利用Adaboost算法对所述正样本Haar特征向量和所述负样本Haar特征向量进行训练,获得第一车牌分类器。S203: Using the Adaboost algorithm to train the positive sample Haar feature vector and the negative sample Haar feature vector to obtain a first license plate classifier.
原始训练集中包括原始正样本和原始负样本,将静态负样本加入原始训练集中后,得到的第一训练集中同样包括正样本和负样本,其中的负样本是指原始负样本与静态负样本的和。The original training set includes original positive samples and original negative samples. After static negative samples are added to the original training set, the obtained first training set also includes positive samples and negative samples. The negative samples refer to the difference between the original negative samples and the static negative samples. and.
基于视频的车牌检测方法有很多,包括基于线模板的二值化图像中的角检测算法,利用遗传算法检测车牌等等。There are many video-based license plate detection methods, including corner detection algorithms in binarized images based on line templates, license plate detection using genetic algorithms, and so on.
在本申请实施例中,优选采用基于Haar特征的Adaboost算法训练第一车牌分类器,使用Haar特征对每一幅车牌的正、负样本进行表征,形成Haar特征向量。最后使用级联的Adaboost算法对Haar特征进行训练,获得第一车牌分类器。In the embodiment of the present application, the first license plate classifier is preferably trained using the Haar feature-based Adaboost algorithm, and the Haar feature is used to characterize the positive and negative samples of each license plate to form a Haar feature vector. Finally, the cascaded Adaboost algorithm is used to train the Haar features to obtain the first license plate classifier.
Haar特征是一种矩形特征,矩形特征对一些简单的图形结构例如边缘,线段比较敏感,但只能描述特定走向,因此比较粗略。但是对于一个检测器,里面包含几十万个不同的矩形特征,再通过使用Adaboost算法进行训练,就可以得到一个强分类器,即第一车牌分类器。The Haar feature is a rectangular feature. The rectangular feature is sensitive to some simple graphic structures such as edges and line segments, but it can only describe a specific direction, so it is relatively rough. But for a detector, which contains hundreds of thousands of different rectangular features, and then trained by using the Adaboost algorithm, a strong classifier can be obtained, that is, the first license plate classifier.
每个Haar的特征模板都是由两个或多个全等的矩形相邻组合而成,特征模板内有白色和黑色两种矩形,并将此模板定义为白色矩形像素和减去黑色矩形像素和。特征模板在子窗口内都可以以任意尺寸任意放置,每一种形态称为一个特征,找出子窗口所有特征,是进行弱分类训练的基础。Each Haar feature template is composed of two or more congruent rectangles adjacent to each other. There are white and black rectangles in the feature template, and this template is defined as white rectangle pixels and black rectangle pixels. and. The feature template can be placed in any size in the sub-window, and each form is called a feature. Finding all the features of the sub-window is the basis for weak classification training.
使用积分图可以加快Haar特征的计算。为了避免一个方框的所有点的边缘值相加的重复计算,在算法中使用了积分图。积分图上的每个点(x,y)包含了从点(0,0)到点(x,y)所有的像素的边缘值。通过使用积分图可以快速得到一个矩形特征中的所有黑色像素的灰度值的和以及所有白色像素的灰度值之和,然后再做一次减法运算即得到一个Haar特征值。Computation of Haar features can be accelerated by using an integral map. In order to avoid repeated calculations of adding up the edge values of all points of a box, an integral map is used in the algorithm. Each point (x,y) on the integral map contains the edge values of all pixels from point (0,0) to point (x,y). By using the integral map, the sum of the gray values of all black pixels and the sum of the gray values of all white pixels in a rectangular feature can be quickly obtained, and then a subtraction operation is performed to obtain a Haar feature value.
Adaboost算法是一种自适应的boosting算法,其基本思想是当分类器对某些样本正确分类时,则减少这些样本的权值。当错误分类时,则增加这些样本的权值,让学习算法在后续的学习中集中对比较难的训练样本进行学习,最终得到一个识别准确率理想的分类器。每一层的训练采用最小允许检测率和最大允许误检率作为强分类器迭代停止依据,当每一层的强分类器的和都达到训练前的设定值时,该级训练即完成。下一层强分类器的训练负样本将从该层中被错误分类的负样本中产生。选择需要使用的Haar特征的类型,载入正样本和负样本。设置虚警率,分类器的层数就可以开始训练。训练过程中每一层训练完毕会测试看是否已经达到了虚警率,如果达到,训练结束。否则一直进行训练直到达到了需要训练的层数。The Adaboost algorithm is an adaptive boosting algorithm, and its basic idea is to reduce the weight of some samples when the classifier correctly classifies some samples. When misclassified, the weights of these samples are increased, so that the learning algorithm concentrates on learning the more difficult training samples in the subsequent learning, and finally obtains a classifier with ideal recognition accuracy. The training of each layer uses the minimum allowable detection rate and the maximum allowable false positive rate as the basis for the iterative stop of the strong classifier. When the sum of the strong classifiers of each layer reaches the set value before training, the training of this level is completed. The training negative samples for the strong classifier in the next layer will be generated from the misclassified negative samples in this layer. Select the type of Haar feature to be used, and load positive and negative samples. Set the false alarm rate, and the number of layers of the classifier can start training. During the training process, after each layer is trained, it will be tested to see if the false alarm rate has been reached. If so, the training ends. Otherwise, keep training until the number of layers that need to be trained is reached.
Adaboost训练流程为:给定一系列的训练样本,初始化每个样本的权重,把权重归一化为一个概率分布,对每个Haar特征训练一个弱分类器,计算对应所有特征的弱分类器的加权和错误率,选取拥有最小错误率的最佳的弱分类器。The Adaboost training process is: Given a series of training samples, initialize the weight of each sample, normalize the weight to a probability distribution, train a weak classifier for each Haar feature, and calculate the weak classifier corresponding to all features Weighting and error rate, select the best weak classifier with the smallest error rate.
训练一个弱分类器(特征f)就是在当前权重分布的情况下,确定f的最优阈值,使得这个弱分类器(特征f)对所有训练样本的分类误差最低。选取一个最佳弱分类器就是选择那个对所有训练样本的分类误差在所有弱分类器中最低的那个弱分类器(特征)。Training a weak classifier (feature f) is to determine the optimal threshold of f under the current weight distribution, so that the weak classifier (feature f) has the lowest classification error for all training samples. Selecting an optimal weak classifier is to select the weak classifier (feature) whose classification error for all training samples is the lowest among all weak classifiers.
第一车牌分类器对待一幅待检测图像时,相当于让所有弱分类器投票,每个弱分类器的权重都不一样,再对投票结果按照弱分类器的错误率加权求和,如果超过了阈值表示当前的样本通过了第一车牌分类器的检测。第一车牌分类器的误检率随着弱分类器数量增多和弱分类器误检率的降低而迅速降低。When the first license plate classifier treats an image to be detected, it is equivalent to asking all weak classifiers to vote. The weight of each weak classifier is different, and then the voting results are weighted and summed according to the error rate of the weak classifier. Exceeding the threshold indicates that the current sample has passed the detection of the first license plate classifier. The false detection rate of the first license plate classifier decreases rapidly as the number of weak classifiers increases and the false detection rate of weak classifiers decreases.
在本申请实施例中,将现场的静态负样本添加到原始训练集,根据得到的第一训练集训练新的第一车牌分类器,训练流程可以为:In the embodiment of the present application, the static negative samples of the scene are added to the original training set, and a new first license plate classifier is trained according to the obtained first training set. The training process can be as follows:
1、使用Haar特征对所述第一训练集里的每一个正样本进行表征,形成正样本Haar特征向量。1. Use the Haar feature to characterize each positive sample in the first training set to form a positive sample Haar feature vector.
2、使用Haar特征对所述第一训练集里的每一个负样本进行表征,形成负样本Haar特征向量。2. Using Haar features to characterize each negative sample in the first training set to form a negative sample Haar feature vector.
3、利用Adaboost算法对1和2中得到的所述正样本Haar特征向量和所述负样本Haar特征向量进行训练,获得第一车牌分类器。3. Using the Adaboost algorithm to train the positive sample Haar feature vector and the negative sample Haar feature vector obtained in 1 and 2, to obtain the first license plate classifier.
本发明使用通过Adaboost算法训练的基于Haar特征的第一车牌分类器进行车牌检测。实践证明,通过Adaboost算法训练的基于Haar特征的第一车牌分类器具有较高的检测率和较低的误检率,并且配合积分图使用,实时性也不存在问题。The present invention uses the Haar feature-based first license plate classifier trained by the Adaboost algorithm to detect the license plate. Practice has proved that the first license plate classifier based on Haar features trained by the Adaboost algorithm has a high detection rate and a low false detection rate, and when used with the integral map, there is no real-time problem.
进一步的,在上述实施例的基础上,根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本的方法,本实施例将结合附图进行说明:Further, on the basis of the above-mentioned embodiments, according to the first license plate classifier to detect on-site car scenes and collect on-site dynamic negative samples, this embodiment will be described in conjunction with the accompanying drawings:
图3为本申请提供的又一种车牌识别方法的流程示意图。FIG. 3 is a schematic flowchart of another license plate recognition method provided by the present application.
请参照图3所示,本申请实施例提供的方法,包括:Please refer to Figure 3, the method provided by the embodiment of the present application includes:
301:获取所述车牌识别设备检测现场的有车场景得到的视频图像,提取所述视频图像中的疑似车牌区域,根据所述第一车牌分类器检测所述疑似车牌区域中的车牌区域;301: Obtain a video image obtained by the license plate recognition device detection scene of a car scene, extract the suspected license plate area in the video image, and detect the license plate area in the suspected license plate area according to the first license plate classifier;
本申请提出的车牌识别方法是粗检加精检的车牌检测策略,即先通过一系列方法,找到疑似包含车牌的区域,然后再使用通过Adaboost算法训练的基于Haar特征的第一车牌分类器对疑似包含车牌的疑似车牌区域进行检测,找到车牌区域。粗检的方法很多,例如提取全图的边缘,寻找边缘密度大的区域。或者寻找全图显著区域等。粗检的细节不在本发明讨论范围。The license plate recognition method proposed in this application is a license plate detection strategy of rough inspection and fine inspection, that is, through a series of methods, find the area suspected to contain the license plate, and then use the first license plate classifier based on Haar features trained by the Adaboost algorithm to pair The suspected license plate area suspected to contain the license plate is detected to find the license plate area. There are many methods of rough inspection, such as extracting the edges of the whole image and looking for areas with high edge density. Or find the salient area of the whole map, etc. The details of the rough check are beyond the scope of this invention.
302:从所述车牌区域中分割出多个字符,根据支持向量机SVM训练的车牌字符识别模型对多个所述字符进行识别,并判断每个所述字符的识别置信度;302: Segment a plurality of characters from the license plate area, recognize the plurality of characters according to the license plate character recognition model trained by the support vector machine SVM, and judge the recognition confidence of each character;
303:根据每个所述字符的识别置信度判断所述车牌区域是否有效,如果无效,将所述车牌区域作为所述现场的动态负样本。303: Determine whether the license plate area is valid according to the recognition confidence of each character, and if not, use the license plate area as a dynamic negative sample of the scene.
在实际使用过程中,对粗检得到的疑似包含车牌区域使用第一车牌分类器进行检测,然后对检测得到的车牌区域进行分割与识别,如果分割与识别的结果不能满足有效车牌的要求,则把该样本作为动态负样本。In the actual use process, the first license plate classifier is used to detect the suspected license plate area obtained by rough inspection, and then the detected license plate area is segmented and recognized. If the segmentation and recognition results cannot meet the requirements of a valid license plate, then Take this sample as a dynamic negative sample.
由于车牌分类器存在一定的漏检和误检,因此检测的结果可能是0个,1个或者多个车牌区域。对检测到的车牌区域进行分割与识别,本发明的字符识别使用的是通过支持向量机训练出来的车牌字符识别模型,然后使用该模型对单个字符进行识别。每个被识别出来的字符都含有一个置信度,最后根据分割和识别的结果判断检测到得目标是否有效:Because the license plate classifier has certain missed detection and false detection, the detection result may be 0, 1 or more license plate areas. The detected license plate area is segmented and recognized. The character recognition of the present invention uses a license plate character recognition model trained by a support vector machine, and then uses the model to recognize a single character. Each recognized character has a confidence level, and finally judges whether the detected target is valid according to the results of segmentation and recognition:
1、车牌必须包含7个字符;1. The license plate must contain 7 characters;
2、每个字符的置信度必须达到各自的阈值(预设的第一阈值);2. The confidence level of each character must reach its respective threshold (preset first threshold);
3、7个字符的置信度加起来的总和也必须达到一个阈值(预设的第二阈值)。3. The sum of the confidence levels of the 7 characters must also reach a threshold (the preset second threshold).
满足以上三个条件的车牌区域即判定为有效车牌,检测出来的车牌区域如果不能满足以上三个条件,就被判定为无效车牌,即动态负样本。The license plate area that meets the above three conditions is judged as a valid license plate, and if the detected license plate area does not meet the above three conditions, it is judged as an invalid license plate, that is, a dynamic negative sample.
另外,在本申请实施例提供的上述方法的基础上,在将动态负样本添加到第一训练集中时,可以包括:In addition, on the basis of the above method provided in the embodiment of the present application, when adding dynamic negative samples to the first training set, it may include:
按照预设的时间间隔统计所述第一车牌分类器检测现场的有车场景的误检率,并判断所述误检率是否大于预设的第三阈值;如果是,判断收集到的所述动态负样本的个数是否大于或等于预设的第四阈值;如果是,将所述动态负样本添加到所述第一训练集中。According to the preset time interval, the false detection rate of the car scene detected by the first license plate classifier is counted, and it is judged whether the false detection rate is greater than the preset third threshold; if so, it is judged that the collected Whether the number of dynamic negative samples is greater than or equal to a preset fourth threshold; if yes, adding the dynamic negative samples to the first training set.
在本申请实施例中,车牌识别设备还可以以一定的时间间隔检查当前系统的误检率。如果误检率大于预设的第三阈值,检查收集到的负样本个数是否满足训练要求(预设的第四阈值),如果不满足,则继续收集动态负样本,如果收集到足够的动态负样本,则把收集的动态负样本添加到第一训练集中,训练第二车牌分类器,并可以持续迭代,用新训练的车牌分类器取代旧的车牌分类器,以进一步降低误检率。In the embodiment of the present application, the license plate recognition device may also check the false detection rate of the current system at a certain time interval. If the false positive rate is greater than the preset third threshold, check whether the number of negative samples collected meets the training requirements (the preset fourth threshold), if not, continue to collect dynamic negative samples, if enough dynamic negative samples are collected Negative samples, add the collected dynamic negative samples to the first training set, train the second license plate classifier, and continue to iterate to replace the old license plate classifier with the newly trained license plate classifier to further reduce the false detection rate.
需要强调的是,本申请实施例中,训练第二车牌分类器使用的第二训练集以及后续迭代过程中的训练集始终包括实验室的原始车牌分类器使用的原始负样本,现场的静态负样本以及现场的动态负样本。It should be emphasized that in the embodiment of this application, the second training set used to train the second license plate classifier and the training set in subsequent iterations always include the original negative samples used by the original license plate classifier in the laboratory, and the static negative samples used in the field. samples as well as live negative samples.
从以上流程可以看出,车牌分类器的训练可以是一个迭代的过程,当达到了重新训练的条件的时候就可以训练一个新的车牌分类器。如果在一定的时间间隔内,车牌识别的效果不好,当达到了满足训练的条件的时候,可以继续进行下一次训练,经过若干次训练后,就能获得一个效果很好,能满足现场环境的新的车牌分类器,而一旦误检率达到了要求,就不需要再进行车牌分类器训练,可以停止迭代。It can be seen from the above process that the training of the license plate classifier can be an iterative process, and a new license plate classifier can be trained when the retraining conditions are met. If the effect of license plate recognition is not good within a certain time interval, when the training conditions are met, you can continue to the next training. The new license plate classifier, and once the false detection rate meets the requirements, there is no need to train the license plate classifier, and the iteration can be stopped.
另外,在本申请实施例中,在所述根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本后,还可以包括:In addition, in the embodiment of the present application, after detecting the car-free scene on the scene according to the original license plate classifier and collecting the static negative samples on the scene, it may also include:
识别所述静态负样本中的简单静态负样本和困难静态负样本,并将所述简单静态负样本添加到所述车牌识别设备的所述原始训练集中;identifying easy static negative samples and difficult static negative samples in the static negative samples, and adding the easy static negative samples to the original training set of the license plate recognition device;
判断所述困难静态负样本所在的图像区域是否满足屏蔽要求;如果满足,屏蔽所述困难静态负样本所在的图像区域;Judging whether the image area where the difficult static negative sample is located meets the masking requirements; if so, shielding the image area where the difficult static negative sample is located;
如果不满足,根据所述困难静态负样本训练第三车牌分类器,所述第三车牌分类器用于判断所述第二车牌分类器识别出的车牌是否为有效车牌。If not, a third license plate classifier is trained according to the difficult static negative samples, and the third license plate classifier is used to judge whether the license plate recognized by the second license plate classifier is a valid license plate.
或者,如果不满足,在预设的时间段内判断所述困难静态负样本所在的图像区域是否发生移动,如果不发生移动,则将所述困难静态负样本添加到所述车牌识别设备的所述原始训练集中。Or, if it is not satisfied, it is judged within the preset time period whether the image area where the difficult static negative sample is located moves, and if there is no movement, the difficult static negative sample is added to all the license plate recognition devices. the original training set.
现场的静态背景会与原始的负样本一起被训练的工程随机挑选。对于现场场景中的简单背景负样本,会被轻易找到一个弱分类器过滤掉,或者在前几层的强分类器就过滤掉。复杂的或者困难的背景的负样本会被一直保留,并且得到权重越来越大,即越来越受关注,很有可能会在较后面的层被过滤掉。The static background of the scene is randomly selected by the training project together with the original negative samples. For simple background negative samples in live scenes, it will be filtered out by easily finding a weak classifier, or by a strong classifier in the first few layers. Negative samples of complex or difficult backgrounds will always be retained, and get more and more weights, that is, more and more attention, and are likely to be filtered out in later layers.
训练结束后使用新的车牌分类器对现场在无车的情况下进行检测。会有一些负样本非常接近车牌,这些负样本即使到了很后的层也无法去除,就必须使用别的方法去除。例如:After training, use the new license plate classifier to detect the scene without a car. There will be some negative samples that are very close to the license plate, and these negative samples cannot be removed even in the last layer, so other methods must be used to remove them. E.g:
1.如果困难的负样本不是出现的图像的中心,或者车辆不会经过,可以把困难的负样本出现的区域屏蔽,即车牌分类器不对该区域进行检测。1. If the difficult negative sample is not the center of the image that appears, or the vehicle will not pass by, the area where the difficult negative sample appears can be shielded, that is, the license plate classifier does not detect the area.
2.如果困难的负样本出现的区域不可以屏蔽,可以专门针对该负样本训练一个分类器,一旦车牌分类器检测到一个车牌就使用该分类器检测判断是否不是车牌。2. If the area where the difficult negative samples appear cannot be masked, you can train a classifier specifically for the negative samples. Once the license plate classifier detects a license plate, use the classifier to detect whether it is not a license plate.
3.可以根据车牌的特性,例如移动性,如果检测到得车牌从第一次被检测到,并且在一定时间内不移动,就认为这个车牌不是有效车牌。例如栅栏有可能无法有效去除,可能识别到得结果全是“1”,“Y”,“T”,“L”等字符,再配合位移信息,也可以认为车牌无效。3. According to the characteristics of the license plate, such as mobility, if the detected license plate is detected for the first time and does not move within a certain period of time, it is considered that the license plate is not a valid license plate. For example, the fence may not be effectively removed, and the recognized results may be all characters such as "1", "Y", "T", "L", etc., combined with the displacement information, the license plate may also be considered invalid.
这样就可以确保新的车牌分类器几乎不会在现场无车的情况下把背景检测成车牌,使用新的车牌分类器进行车牌检测。This ensures that the new license plate classifier will hardly detect the background as a license plate when there is no car on the scene, and use the new license plate classifier for license plate detection.
针对现场不能去除的困难静态样本,本发明提出通过添加屏蔽区域,训练特殊分类器以及结合位移信息等方法进行去除。动态误检由于不可预知,去除比较困难,但是如果能有效去除静态误检,并尽可能减少动态误检,就能达到较低的误检率。For difficult static samples that cannot be removed on site, the present invention proposes to remove them by adding shielding areas, training special classifiers, and combining displacement information. Due to the unpredictability of dynamic false detection, it is more difficult to remove, but if the static false detection can be effectively removed and the dynamic false detection can be reduced as much as possible, a lower false detection rate can be achieved.
上述实施例为本申请提供的方法实施例,对应上述方法实施例,本申请还提供一种车牌识别设备。The above-mentioned embodiments are method embodiments provided by the present application, and corresponding to the above-mentioned method embodiments, the present application also provides a license plate recognition device.
图4为本申请提供的一种车牌识别设备的结构示意图。FIG. 4 is a schematic structural diagram of a license plate recognition device provided by the present application.
请参照图4所示,本申请实施例提供的车牌识别设备,用于布置在现场进行车牌识别,所述车牌识别设备中包括根据原始训练集训练的原始车牌分类器,所述车牌识别设备还包括:Please refer to Fig. 4, the license plate recognition device provided by the embodiment of the present application is used to arrange on-site license plate recognition, the license plate recognition device includes an original license plate classifier trained according to the original training set, and the license plate recognition device also include:
静态负样本收集模块1,用于根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本;所述静态负样本为所述现场的无车场景中被所述原始车牌分类器误检为车牌区域的背景图像;Static negative sample collection module 1, for detecting the car-free scene of the scene according to the original license plate classifier, collect the static negative sample of the scene; The static negative sample is classified by the original license plate in the car-free scene of the scene The device is falsely detected as the background image of the license plate area;
第一车牌分类器模块2,用于将所述静态负样本添加到所述车牌识别设备的所述原始训练集中,得到第一训练集,根据所述第一训练集训练第一车牌分类器;The first license plate classifier module 2 is used to add the static negative sample to the original training set of the license plate recognition device to obtain a first training set, and train the first license plate classifier according to the first training set;
第二车牌分类器模块3,用于根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本,将所述动态负样本添加到所述第一训练集中,得到第二训练集,根据所述第二训练集训练第二车牌分类器,并根据所述第二车牌分类器进行车牌识别;所述动态负样本为所述现场的有车场景中被所述第一车牌分类器误检为车牌区域的运动图像。The second license plate classifier module 3 is used to detect the car scene on the scene according to the first license plate classifier, collect the dynamic negative samples of the scene, add the dynamic negative samples to the first training set, and obtain the second license plate classifier. training set, training a second license plate classifier according to the second training set, and performing license plate recognition according to the second license plate classifier; The classifier misdetects moving images as license plate regions.
对应于上述方法实施例,本实施例提供的车牌识别设备所采用的车牌识别方法和识别原理与上述方法实施例类似,此处不再赘述。Corresponding to the above method embodiments, the license plate recognition method and recognition principle adopted by the license plate recognition device provided in this embodiment are similar to the above method embodiments, and will not be repeated here.
同时,在上述实施例的基础上,在本申请实施例中,At the same time, on the basis of the above embodiments, in the embodiments of this application,
所述第一车牌分类器模块2,可以包括:The first license plate classifier module 2 may include:
正样本特征向量单元,用于使用Haar特征对所述第一训练集里的每一个正样本进行表征,形成正样本Haar特征向量;A positive sample feature vector unit, configured to characterize each positive sample in the first training set using Haar features to form a positive sample Haar feature vector;
负样本特征向量单元,用于使用Haar特征对所述第一训练集里的每一个负样本进行表征,形成负样本Haar特征向量;Negative sample feature vector unit, for using Haar feature to represent each negative sample in the first training set to form a negative sample Haar feature vector;
训练单元,用于利用Adaboost算法对所述正样本Haar特征向量和所述负样本Haar特征向量进行训练,获得第一车牌分类器。The training unit is configured to use the Adaboost algorithm to train the positive sample Haar feature vector and the negative sample Haar feature vector to obtain a first license plate classifier.
所述第二车牌分类器模块3,可以包括:The second license plate classifier module 3 may include:
检测单元,用于获取所述车牌识别设备检测现场的有车场景得到的视频图像,提取所述视频图像中的疑似车牌区域,根据所述第一车牌分类器检测所述疑似车牌区域中的车牌区域;A detection unit, configured to acquire a video image obtained by the vehicle scene detected by the license plate recognition device, extract the suspected license plate area in the video image, and detect the license plate in the suspected license plate area according to the first license plate classifier area;
识别单元,用于从所述车牌区域中分割出多个字符,根据支持向量机SVM训练的车牌字符识别模型对多个所述字符进行识别,并判断每个所述字符的识别置信度;A recognition unit, configured to segment a plurality of characters from the license plate area, recognize a plurality of the characters according to the license plate character recognition model trained by the support vector machine SVM, and judge the recognition confidence of each of the characters;
动态负样本单元,用于根据每个所述字符的识别置信度判断所述车牌区域是否有效,如果无效,将所述车牌区域作为所述现场的动态负样本。The dynamic negative sample unit is used to judge whether the license plate area is valid according to the recognition confidence of each character, and if invalid, use the license plate area as a dynamic negative sample of the scene.
所述动态负样本单元,可以包括:The dynamic negative sample unit may include:
第一判断子单元,用于判断所述字符的个数是否为7个;The first judging subunit is used to judge whether the number of the characters is 7;
第二判断子单元,用于如果是,判断每个所述字符的识别置信度是否大于或等于第一阈值;The second judging subunit is configured to, if yes, judging whether the recognition confidence of each character is greater than or equal to the first threshold;
第三判断子单元,用于如果是,判断7个所述字符的识别置信度的和是否大于或等于第二阈值;The third judging subunit is used to judge whether the sum of the recognition confidences of the 7 characters is greater than or equal to the second threshold;
判定子单元,用于如果是,则判定所述车牌区域有效,否则无效。A judging subunit, configured to judge that the license plate area is valid if yes, otherwise it is invalid.
所述第二车牌分类器模块3,可以包括:The second license plate classifier module 3 may include:
统计单元,用于按照预设的时间间隔统计所述第一车牌分类器检测现场的有车场景的误检率,并判断所述误检率是否大于第三阈值;A statistical unit, configured to count the false detection rate of the car scene detected by the first license plate classifier according to a preset time interval, and determine whether the false detection rate is greater than a third threshold;
判断单元,用于如果是,判断收集到的所述动态负样本的个数是否大于或等于第四阈值;a judging unit, configured to, if yes, judge whether the number of the collected dynamic negative samples is greater than or equal to a fourth threshold;
添加单元,用于如果是,将所述动态负样本添加到所述第一训练集中。The adding unit is configured to, if yes, add the dynamic negative samples to the first training set.
所述静态负样本收集模块1,包括:The static negative sample collection module 1 includes:
识别单元,用于识别所述静态负样本中的简单静态负样本和困难静态负样本,并将所述简单静态负样本添加到所述车牌识别设备的所述原始训练集中;A recognition unit, configured to identify simple static negative samples and difficult static negative samples in the static negative samples, and add the simple static negative samples to the original training set of the license plate recognition device;
判断单元,用于判断所述困难静态负样本所在的图像区域是否满足屏蔽要求;A judging unit, configured to judge whether the image region where the difficult static negative sample is located satisfies the masking requirement;
屏蔽单元,用于如果满足,屏蔽所述困难静态负样本所在的图像区域,如果不满足,根据所述困难静态负样本训练第三车牌分类器,所述第三车牌分类器用于判断所述第二车牌分类器识别出的车牌是否为有效车牌。The shielding unit is used to shield the image area where the difficult static negative sample is located if it is satisfied, and if it is not satisfied, train a third license plate classifier according to the difficult static negative sample, and the third license plate classifier is used to judge the first Second, whether the license plate recognized by the license plate classifier is a valid license plate.
所述静态负样本收集模块1,包括:The static negative sample collection module 1 includes:
识别单元,用于识别所述静态负样本中的简单静态负样本和困难静态负样本,并将所述简单静态负样本添加到所述车牌识别设备的所述原始训练集中;A recognition unit, configured to identify simple static negative samples and difficult static negative samples in the static negative samples, and add the simple static negative samples to the original training set of the license plate recognition device;
判断单元,用于判断所述困难静态负样本所在的图像区域是否满足屏蔽要求;A judging unit, configured to judge whether the image region where the difficult static negative sample is located satisfies the masking requirement;
屏蔽单元,用于如果满足,屏蔽所述困难静态负样本所在的图像区域,如果不满足,在预设的时间段内判断所述困难静态负样本所在的图像区域是否发生移动,如果不发生移动,则将所述困难静态负样本添加到所述车牌识别设备的所述原始训练集中。The shielding unit is used to shield the image area where the difficult static negative sample is located if it is satisfied, and if it is not satisfied, judge whether the image area where the difficult static negative sample is located moves within a preset time period, and if it does not move , then add the difficult static negative samples to the original training set of the license plate recognition device.
对应于上述方法实施例,本实施例提供的车牌识别设备所采用的车牌识别方法和识别原理与上述方法实施例类似,此处不再赘述。Corresponding to the above method embodiments, the license plate recognition method and recognition principle adopted by the license plate recognition device provided in this embodiment are similar to the above method embodiments, and will not be repeated here.
由以上本申请提供的技术方案,车牌识别方法应用于布置在现场的车牌识别设备中,所述车牌识别设备中包括根据原始训练集训练的原始车牌分类器,所述车牌识别方法根据所述原始车牌分类器检测现场的无车场景,收集现场的静态负样本;所述静态负样本为所述现场的无车场景中被所述原始车牌分类器误检为车牌区域的背景图像;将所述静态负样本添加到所述车牌识别设备的所述原始训练集中,得到第一训练集,根据所述第一训练集训练第一车牌分类器;根据所述第一车牌分类器检测现场的有车场景,收集现场的动态负样本,将所述动态负样本添加到所述第一训练集中,得到第二训练集,根据所述第二训练集训练第二车牌分类器,并根据所述第二车牌分类器进行车牌识别;所述动态负样本为所述现场的有车场景中被所述第一车牌分类器误检为车牌区域的运动图像。这样,先收集现场的静态负样本并添加到原始训练集,得到第一训练集,利用第一训练集训练第一车牌分类器去除静态负样本,然后再使用第一车牌分类器收集动态负样本并添加到第一训练集,得到第二训练集,利用第二训练集训练第二车牌分类器就可以去除收集到的静态负样本和动态负样本,此时第二训练集中已经包含了现场可能出现的绝大部分负样本,这些负样本能够被第二车牌分类器拒绝,从而达到对车牌的精确识别,减少车牌识别设备的误检率。Based on the above technical solution provided by the present application, the license plate recognition method is applied to the license plate recognition equipment arranged on site, and the license plate recognition equipment includes an original license plate classifier trained according to the original training set, and the license plate recognition method is based on the original The license plate classifier detects the car-free scene of the scene, and collects the static negative samples of the scene; the static negative sample is the background image of the license plate area misdetected by the original license plate classifier in the car-free scene of the scene; Static negative samples are added to the original training set of the license plate recognition device to obtain a first training set, and a first license plate classifier is trained according to the first training set; Scenario, collect the dynamic negative samples of the scene, add the dynamic negative samples to the first training set to obtain the second training set, train the second license plate classifier according to the second training set, and according to the second training set The license plate classifier performs license plate recognition; the dynamic negative sample is a moving image in the on-site vehicle scene that is misdetected as a license plate area by the first license plate classifier. In this way, first collect the static negative samples on the scene and add them to the original training set to obtain the first training set, use the first training set to train the first license plate classifier to remove static negative samples, and then use the first license plate classifier to collect dynamic negative samples And add it to the first training set to get the second training set, and use the second training set to train the second license plate classifier to remove the collected static negative samples and dynamic negative samples. At this time, the second training set already contains the on-site possible Most of the negative samples that appear, these negative samples can be rejected by the second license plate classifier, so as to achieve accurate recognition of the license plate and reduce the false detection rate of the license plate recognition device.
需要说明的是,本说明书中的各个实施例均采用递进的方式描述,每个实施例重点说明的都是与其他实施例的不同之处,各个实施例之间相同相似的部分互相参见即可。对于装置类实施例而言,由于其与方法实施例基本相似,所以描述的比较简单,相关之处参见方法实施例的部分说明即可。It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. For the same and similar parts in each embodiment, refer to each other, that is, Can. As for the device-type embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, please refer to part of the description of the method embodiments.
最后,还需要说明的是,在本文中,诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来,而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。Finally, it should also be noted that in this text, relational terms such as first and second etc. are only used to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations, any such actual relationship or order exists. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.
以上对本发明所提供的一种车牌识别方法及车牌识别设备进行了详细介绍,本文中应用了具体个例对本发明的原理及实施方式进行了阐述,以上实施例的说明只是用于帮助理解本发明的方法及其核心思想;同时,对于本领域的一般技术人员,依据本发明的思想,在具体实施方式及应用范围上均会有改变之处,综上所述,本说明书内容不应理解为对本发明的限制。The above is a detailed introduction to a license plate recognition method and license plate recognition device provided by the present invention. In this paper, specific examples are used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only used to help understand the present invention. method and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation and scope of application. In summary, the content of this specification should not be understood as Limitations on the Invention.
Claims (10)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410742690.8A CN104361366B (en) | 2014-12-08 | 2014-12-08 | License plate recognition method and license plate recognition device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410742690.8A CN104361366B (en) | 2014-12-08 | 2014-12-08 | License plate recognition method and license plate recognition device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104361366A CN104361366A (en) | 2015-02-18 |
CN104361366B true CN104361366B (en) | 2018-10-30 |
Family
ID=52528624
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410742690.8A Expired - Fee Related CN104361366B (en) | 2014-12-08 | 2014-12-08 | License plate recognition method and license plate recognition device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104361366B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106295635A (en) * | 2016-02-17 | 2017-01-04 | 北京智芯原动科技有限公司 | Licence plate recognition method and system |
CN107016356A (en) * | 2017-03-21 | 2017-08-04 | 乐蜜科技有限公司 | Certain content recognition methods, device and electronic equipment |
CN112149707B (en) * | 2019-06-28 | 2024-06-14 | 商汤集团有限公司 | Image acquisition control method, device, medium and equipment |
CN111310850B (en) * | 2020-03-02 | 2023-06-16 | 杭州雄迈集成电路技术股份有限公司 | License plate detection model construction method and system, license plate detection method and system |
CN113496227A (en) * | 2020-04-08 | 2021-10-12 | 顺丰科技有限公司 | Training method and device of character recognition model, server and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101520841A (en) * | 2009-03-10 | 2009-09-02 | 北京航空航天大学 | Real-time and anti-interference method for positioning license plate in high-definition TV video |
CN103390156A (en) * | 2012-11-05 | 2013-11-13 | 深圳市捷顺科技实业股份有限公司 | License plate recognition method and device |
KR20140093407A (en) * | 2013-01-18 | 2014-07-28 | 광주과학기술원 | Recognition device, vehicle model recognition apparatus and method |
CN104408475A (en) * | 2014-12-08 | 2015-03-11 | 深圳市捷顺科技实业股份有限公司 | License plate recognition method and license plate recognition device |
-
2014
- 2014-12-08 CN CN201410742690.8A patent/CN104361366B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101520841A (en) * | 2009-03-10 | 2009-09-02 | 北京航空航天大学 | Real-time and anti-interference method for positioning license plate in high-definition TV video |
CN103390156A (en) * | 2012-11-05 | 2013-11-13 | 深圳市捷顺科技实业股份有限公司 | License plate recognition method and device |
KR20140093407A (en) * | 2013-01-18 | 2014-07-28 | 광주과학기술원 | Recognition device, vehicle model recognition apparatus and method |
CN104408475A (en) * | 2014-12-08 | 2015-03-11 | 深圳市捷顺科技实业股份有限公司 | License plate recognition method and license plate recognition device |
Non-Patent Citations (1)
Title |
---|
基于Adaboost算法的驾驶员眨眼识别;赵雪竹等;《电子产品世界》;20090904;第31-32、42页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104361366A (en) | 2015-02-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104408475B (en) | License plate recognition method and license plate recognition equipment | |
CN109993056B (en) | Method, server and storage medium for identifying vehicle illegal behaviors | |
CN111563494B (en) | Behavior identification method and device based on target detection and computer equipment | |
Abedin et al. | License plate recognition system based on contour properties and deep learning model | |
CN105930791B (en) | Recognition method of road traffic signs based on multi-camera fusion based on DS evidence theory | |
Jiao et al. | A configurable method for multi-style license plate recognition | |
Puranic et al. | Vehicle number plate recognition system: a literature review and implementation using template matching | |
CN104361366B (en) | License plate recognition method and license plate recognition device | |
CN103871077B (en) | A kind of extraction method of key frame in road vehicles monitoring video | |
CN102346847A (en) | License plate character recognizing method of support vector machine | |
Prates et al. | Brazilian license plate detection using histogram of oriented gradients and sliding windows | |
Akbarzadeh et al. | Design and matlab simulation of Persian license plate recognition using neural network and image filtering for intelligent transportation systems | |
CN109993134A (en) | A vehicle detection method at road intersection based on HOG and SVM classifier | |
Prabhu et al. | Recognition of Indian license plate number from live stream videos | |
Panem et al. | Next-Gen Smart Vehicle Number Plate Recognition with Dynamic Character Encoding for Lightning-Fast and Accurate Detection | |
Prajapati et al. | A review paper on automatic number plate recognition using machine learning: An in-depth analysis of machine learning techniques in automatic number plate recognition: Opportunities and limitations | |
Zhang et al. | Image-based approach for parking-spot detection with occlusion handling | |
Elbamby et al. | Real-time automatic multi-style license plate detection in videos | |
Al Nasim et al. | An automated approach for the recognition of bengali license plates | |
Yuhana et al. | Enhancement of blurred Indonesian license plate number identification using multi-scale information CNN | |
CN110619255B (en) | Target detection method and device | |
Kim et al. | Effective traffic lights recognition method for real time driving assistance system in the daytime | |
CN104331708B (en) | A kind of zebra crossing automatic detection analysis method and system | |
Amin et al. | An automatic number plate recognition of Bangladeshi vehicles | |
Kodwani et al. | Automatic license plate recognition in real time videos using visual surveillance techniques |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20181030 |