CN104077568A - High-accuracy driver behavior recognition and monitoring method and system - Google Patents

Abstract

The invention discloses a high-precision driver behavior recognition and monitoring method and system. The method includes: acquiring real-time video frames through dual cameras; performing target part detection and target part feature extraction on the acquired real-time video frames, thereby obtaining real-time targets Part feature points; compare the real-time target part feature points with the standard target part feature points obtained according to the training samples and target templates, and identify and analyze the driver's driving behavior according to the comparison results; identify and analyze the Results are fed back to the user and displayed in real time. The method of the present invention solves the non-real-time deficiencies of the prior art through the real-time video data collected by dual cameras; adopts dual cameras to collect real-time video images, which is more accurate; detects target parts, extracts features of target parts and compares feature points Yes, the accuracy of identification and monitoring has been further improved. The invention can be widely used in the fields of intelligent transportation and image processing.

Description

A high-precision driver behavior recognition and monitoring method and system

technical field

The invention relates to the technical field of intelligent transportation and image processing, in particular to a high-precision driver behavior recognition and monitoring method and system.

Background technique

With the increase in the number of cars and the expansion of road construction scale, problems such as traffic accidents are becoming more and more obvious. China is the country with the largest population and the country with the highest number of road traffic accident deaths, ranking first in the world for several consecutive years. The vast majority of traffic accidents are caused by driver error and fatigue driving. Due to changes in age, physical or mental health, emotions, etc., even a good driver may not be able to maintain his original good driving state for a long time, but it is difficult for the driver himself to realize this gradual attenuation or subside. Therefore, identifying and monitoring the driver's driving behavior and giving an alert to violations can improve the driver's driving ability and reduce his driving load, coordinate the relationship between the driver, the vehicle and the traffic environment, and essentially reduce traffic accidents The occurrence of the situation is of great significance.

At present, many research results have been obtained in identifying and monitoring driver behavior at home and abroad, which can be roughly divided into two methods: subjective and objective. Among them, subjective research methods include subjective questionnaires and driver self-records. Objective methods include measurements such as electroencephalography, electrooculogram, electromyography, respiratory airflow, respiratory effect, temperature at oxygen saturation of arterial blood, and electrocardiogram. Although the above methods are relatively accurate, these methods are leading or lagging rather than real-time. Generally, they are measured before or after driving, and the real-time performance is poor, and complex detection instruments need to be placed in the limited space of the cab. , the installation is difficult. In addition, the existing technology generally only uses a single camera to obtain driving images, and there are dead spots in the video, and the driving images cannot be fully collected, and the accuracy is low.

To sum up, the industry urgently needs a real-time, low installation difficulty, high-precision driver behavior recognition and monitoring method and system.

Contents of the invention

In order to solve the above-mentioned technical problems, the object of the present invention is to provide a real-time, low installation difficulty and high-precision driver behavior recognition and monitoring system.

Another object of the present invention is to provide a real-time, low-installation difficulty, and high-precision driver behavior recognition and monitoring method.

The technical solution adopted by the present invention to solve the technical problem is: a high-precision driver behavior recognition and monitoring system, comprising:

Dual cameras for collecting real-time video data frames;

The video decoder is used to decode the data collected by the dual cameras to obtain real-time video image data frames;

Hard disk recorder, used to store real-time video image data frames, and upload real-time video image data frames to the industrial computer processing center;

The industrial computer processing center is used to perform feature detection, feature extraction and real-time state analysis based on the uploaded real-time video image data frame, so as to identify and monitor the driver's behavior;

Alarm module, used for alarming abnormal driving behavior;

The dual camera is connected with the input terminal of the alarm module through the video decoder, the hard disk recorder and the industrial computer processing center in turn.

Further, the industrial computer processing center includes:

The target part detection unit is used to detect the target part from the real-time video image data frame;

The target part high-precision positioning unit is used to extract the target part features from the candidate area;

The driving state identification and analysis unit is used to compare the features of the target part with the standard features, and identify and analyze the driving behavior and state of the driver according to the comparison results;

The input end of the target position detection unit is connected with the output end of the hard disk recorder, the output end of the target position detection unit is further connected with the input end of the driving state recognition and analysis unit through the target position high-precision positioning unit, the driving state The output end of the state recognition and analysis unit is connected with the input end of the alarm module.

Further, it also includes an on-board power supply for high-resolution dual cameras and a power supply module for hard disk recorders and industrial computer processing centers.

Another technical solution adopted by the present invention to solve the technical problem is: a high-precision driver behavior recognition and monitoring method, including:

A. Obtain real-time video frames through dual cameras;

B. Perform target part detection and target part feature extraction on the acquired real-time video frame, so as to obtain real-time target part feature points;

C. Compare the real-time feature points of the target part with the standard target part feature points obtained according to the training samples and target templates, and identify and analyze the driver's driving behavior according to the comparison results;

D. Feedback the identification and analysis results to the user and display them in real time.

Further, the target parts include head and human face.

Further, the step B includes:

B1. Perform face detection on the acquired video frame according to preset screening conditions, thereby detecting a human face image, and the preset screening conditions are:

Among them, m is the main function, K is the objective function, τ is the visible coefficient, is the area coefficient, is the detected scale factor;

B2. Perform eye positioning, normalization processing and feature point extraction on the face image, so as to extract real-time face feature points.

Further, in the step B2, the step of eye positioning is performed on the face image, which is specifically:

Perform eye positioning on the face image according to a preset feature point calculation formula, thereby locating the eye region, and the preset feature point calculation formula is:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span><span\; class="notranslate">\; \&infin;</span>}{\mathrm{<span\; class="notranslate">\; lim</span>}}<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span><span\; class="notranslate">\; \&infin;</span>}{\mathrm{<span\; class="notranslate">\; lim</span>}}<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>$

Among them, E is the unit pixel point, L is the area of the pixel point x, D, S, A, N are the four corner points of the pixel point x, and h is the target point that may be located.

Further, the step of normalizing the face image in the step B2 is specifically:

The eye area is normalized according to the distance between the two eyes, so as to obtain the effective face area, and the effective face area satisfies:

Among them, k is the radius of the pupil of the eye, d is the distance between the two eyes, is the pixel coefficient, is the photometric coefficient, V and D are the coordinate points between the two eyes, and L is the main function.

Further, the step of performing normalization processing on the face image in the step B2 also includes the step of image enhancement on the face image, and the formula used in the step of image enhancement is:

$\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; min</span>}}\underset{\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; \&tau;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; \&chi;</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; \&tau;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; \&chi;</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>$

Among them, j is the value range from the center of the face to the center point d, θ is the value range from the center of the mouth to the detection point R, D is the original image pixel, K is the target pixel function after processing, χ _τ , j is the pixel in the processing area.

Further, said step C, which includes:

C1, comparing the real-time target part feature points with the standard target part feature points obtained according to the training sample and the target template, so as to identify the video target image, the formula based on which the video target image is identified is:

Among them, α is the pixel coefficient, K is a fixed value constant, χ _τ are the abscissas of the video target image, υ _j are the ordinates of the trackable coefficients of the video target image, γ is the gray scale coefficient, and i, j are the sequence coefficients;

C2. Analyze the driver's driving behavior according to the recognition result.

The beneficial effects of the system of the present invention are: through the processing center of the industrial computer, the detection, identification and monitoring of the driver's driving behavior can be realized in a completely natural and non-contact manner, combined with the real-time video collection results, the real-time Data, analyzing the real-time video frame, solves the lack of non-real-time performance of the existing technology; the required detection instruments are small enough, which reduces the difficulty of installing hardware equipment and saves cab space; dual cameras are used to collect real-time video The image overcomes the shortcomings of only using a single camera to collect images, and can comprehensively collect driving images, which is more accurate.

The beneficial effects of the method of the present invention are: the real-time video data collected by the dual cameras can quickly and flexibly acquire the real-time data, and analyze the real-time video frames, which solves the non-real-time deficiency of the prior art; Collecting real-time video images overcomes the shortcomings of only using a single camera to collect images, and can comprehensively collect driving images, which is more accurate; through target part detection, target part feature extraction and feature point comparison, it is possible to analyze the driver's facial expressions, Real-time recognition of facial features, eye movement rules, eye status, and head movement rules, which further improves the accuracy of recognition and monitoring.

Description of drawings

The present invention will be further described below in conjunction with drawings and embodiments.

Fig. 1 is a functional block diagram of a high-precision driver behavior recognition and monitoring system of the present invention;

Fig. 2 is the structural block diagram of industrial computer processing center of the present invention;

Fig. 3 is a flow chart of the steps of a high-precision driver behavior recognition and monitoring method of the present invention;

Fig. 4 is the flowchart of step B of the present invention;

Fig. 5 is a flowchart of step C of the present invention.

Detailed ways

Referring to Figure 1, a high-precision driver behavior recognition and monitoring system includes:

Dual cameras for collecting real-time video data frames;

The video decoder is used to decode the data collected by the dual cameras to obtain real-time video image data frames;

Hard disk recorder, used to store real-time video image data frames, and upload real-time video image data frames to the industrial computer processing center;

The industrial computer processing center is used to perform feature detection, feature extraction and real-time state analysis based on the uploaded real-time video image data frame, so as to identify and monitor the driver's behavior;

Alarm module, used for alarming abnormal driving behavior;

The dual cameras are connected with the input terminal of the alarm module through the video decoder, the hard disk recorder and the industrial computer processing center in turn.

Referring to Fig. 2, further as a preferred embodiment, the industrial computer processing center includes:

The target part detection unit is used to detect the target part from the real-time video image data frame;

The target part high-precision positioning unit is used to extract the target part features from the candidate area;

The driving state identification and analysis unit is used to compare the features of the target part with the standard features, and identify and analyze the driving behavior and state of the driver according to the comparison results;

The input end of the target position detection unit is connected with the output end of the hard disk recorder, the output end of the target position detection unit is further connected with the input end of the driving state recognition and analysis unit through the target position high-precision positioning unit, the driving state The output end of the state recognition and analysis unit is connected with the input end of the alarm module.

Wherein, the target parts include human face and head.

Referring to FIG. 1 , as a further preferred embodiment, it also includes an on-board power supply for the high-resolution dual cameras and a power supply module for the hard disk recorder and the industrial computer processing center.

Referring to Figure 3, a high-precision driver behavior recognition and monitoring method includes:

A. Obtain real-time video frames through dual cameras;

B. Perform target part detection and target part feature extraction on the acquired real-time video frame, so as to obtain real-time target part feature points;

C. Compare the real-time feature points of the target part with the standard target part feature points obtained according to the training samples and target templates, and identify and analyze the driver's driving behavior according to the comparison results:

D. Feedback the identification and analysis results to the user and display them in real time.

As a further preferred embodiment, the target site includes a head and a human face.

Referring to Fig. 4, further as a preferred embodiment, the step B includes:

B1. Perform face detection on the acquired video frame according to preset screening conditions, thereby detecting a human face image, and the preset screening conditions are:

Among them, m is the main function, K is the objective function, τ is the visible coefficient, is the area coefficient, is the detected scale factor;

B2. Perform eye positioning, normalization processing and feature point extraction on the face image, so as to extract real-time face feature points.

Wherein, the human face detection of the present invention is completely consistent with the head detection process, and the effect of the human face detection process is greater.

Normalization processing includes operations such as image preprocessing, image scaling, and effective face area selection.

Further as a preferred embodiment, in the step B2, the step of eye positioning is performed on the face image, which is specifically:

Perform eye positioning on the face image according to a preset feature point calculation formula, thereby locating the eye region, and the preset feature point calculation formula is:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span><span\; class="notranslate">\; \&infin;</span>}{\mathrm{<span\; class="notranslate">\; lim</span>}}<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span><span\; class="notranslate">\; \&infin;</span>}{\mathrm{<span\; class="notranslate">\; lim</span>}}<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>$

Among them, E is the unit pixel point, L is the area of the pixel point x, D, S, A, N are the four corner points of the pixel point x, and h is the target point that may be located.

Further as a preferred embodiment, the step of normalizing the face image in the step B2 is specifically:

The eye area is normalized according to the distance between the two eyes, so as to obtain the effective face area, and the effective face area satisfies:

Among them, k is the radius of the pupil of the eye, d is the distance between the two eyes, is the pixel coefficient, is the photometric coefficient, V and D are the coordinate points between the two eyes, and L is the main function.

Further as a preferred embodiment, the step of performing normalization processing on the face image in the step B2 also includes the step of image enhancement on the face image, and the formula used in the step of image enhancement is:

$\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; min</span>}}\underset{\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; \&tau;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; \&chi;</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; \&tau;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; \&chi;</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>$

Among them, j is the value range from the center of the face to the center point d, θ is the value range from the center of the mouth to the detection point R, D is the original image pixel, K is the target pixel function after processing, χ _τ , j is the pixel in the processing area.

Referring to Fig. 5, further as a preferred embodiment, the step C includes:

C1, comparing the real-time target part feature points with the standard target part feature points obtained according to the training sample and the target template, so as to identify the video target image, the formula based on which the video target image is identified is:

Among them, α is the pixel coefficient, K is a fixed value constant, χ _τ are the abscissas of the video target image, υ _j are the ordinates of the trackable coefficients of the video target image, γ is the gray scale coefficient, and i, j are the sequence coefficients;

C2. Analyze the driver's driving behavior according to the recognition result.

The present invention will be described in further detail below in conjunction with specific embodiments.

Embodiment one

Referring to Fig. 1, the first embodiment of the present invention:

The present invention adopts double cameras to collect video, can realize 3D effect, can make up for the deficiency of a single camera, and achieves a comprehensive and low-error capture effect. Among them, the dual camera can quickly capture real-time video. The driver's behavior recognition and monitoring system of the present invention is mainly composed of two parts, a video acquisition module and an industrial computer processing center, wherein the video acquisition module includes dual cameras, a video decoder and a hard disk recorder, and the whole system structure is as shown in Figure 1. Among them, the selection of the two cameras is unified, and its device interface provides a variety of real-time data transmission modes. Each mode uses data packets of different sizes, and the speed can reach up to 7.5Mbps. It has 16 dynamic memories of 256k and supports 16 bit or 8-bit YUV 4:2:2 or 4:1:1RGB data format. When the system is working, the dual cameras collect real-time video data frames, which are processed and stored in the video storage space of the hard disk recorder, and then displayed through the display video window of the industrial computer processing center, and the video is analyzed and tracked according to the target requirements , and finally conduct state analysis and reflect the analysis results.

Embodiment two

This embodiment describes the process of preprocessing a face image.

Considering the feasibility of data collection in real life, in order to ensure the consistency of face size, position and face image quality in the face image, the present invention needs to perform data preprocessing on the image before face feature extraction. The main purpose of image data preprocessing is to eliminate irrelevant information in the image, filter out interference and noise, restore useful real information, enhance the detectability of relevant information and simplify data to the greatest extent, thereby improving feature extraction, image segmentation, and matching. and recognition reliability.

The preprocessing process of the human face image in the present invention mainly includes processes such as face straightening, enhancement and normalization of the human face image. Face straightening is to obtain a face image with a correct face position, and image enhancement is to improve the quality of the face image, not only to make the image clearer visually, but also to make the image more conducive to computer processing and recognition. Among them, image enhancement uses the following formula:

$\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; min</span>}}\underset{\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; \&tau;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; \&chi;</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; \&tau;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; \&chi;</span>}}_{\mathrm{<span\; class="notranslate">\; \&tau;</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; .</span>$

The goal of normalization work is to obtain standardized face images with the same size and the same gray value range.

Embodiment Three

This embodiment introduces the process of generating training samples and target templates of the present invention.

The face detection and recognition process of the present invention adopts a classifier selection method based on AdaBoost, and a feature set and a training set including positive and negative sample images are given. Any machine learning method can be used to train the classification function of the present invention through learning. Therefore, the present invention uses the AdaBoost method to perform sample training, select features and train classifiers, thereby generating training samples and target templates for extracting standard feature points.

Compared with the prior art, the present invention can detect, recognize and monitor the driver's driving behavior in a completely natural and non-contact manner through the processing center of the industrial computer, and combined with the real-time video collection results, the real-time data can be obtained quickly and flexibly , to analyze the real-time video frame, which solves the non-real-time deficiency of the existing technology; the required detection instruments are less enough, which reduces the difficulty of installing hardware equipment and saves cab space; adopts dual cameras to collect real-time video images , overcome the shortcomings of only using a single camera to collect images, and can comprehensively collect driving images, which is more accurate; through target part detection, target part feature extraction and feature point comparison, it is possible to analyze the driver's facial expressions, facial features, and eyes. Real-time recognition of movement rules, eye status, head movement rules, etc., further improving the accuracy of recognition and monitoring.

The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the described embodiments, and those skilled in the art can also make various equivalent deformations or replacements without violating the spirit of the present invention. , these equivalent modifications or replacements are all within the scope defined by the claims of the present application.

Claims (10)

1. high-precision driving behavior is identified and a supervisory system, it is characterized in that comprising:

Dual camera, for gathering real time video data frame;

Video Decoder, for the decoding data that dual camera is gathered, thereby obtains real time video image Frame;

Hard disk recorder, for storing real time video image Frame, and is uploaded to industrial computer processing enter by real time video image Frame;

Industrial computer processing enter, for carrying out feature detection, feature extraction and real-time status analysis according to the real time video image Frame of uploading, thereby identifies and monitoring driving behavior;

Alarm modules, for to abnormal driving behavior report to the police;

Described dual camera is connected by Video Decoder, hard disk recorder and industrial computer processing enter and then with the input end of alarm modules successively.

2. a kind of high-precision driving behavior according to claim 1 is identified and supervisory system, it is characterized in that: described industrial computer processing enter comprises:

Target site detecting unit, for detecting target site from real time video image Frame;

The high-precision fixed bit location of target site, for extracting target site feature from candidate region;

Driving condition identification and analysis unit, for target site feature and standard feature are compared, and carries out identification and analysis according to the result of comparison to driver's driving behavior and state;

The input end of described target site detecting unit is connected with the output terminal of hard disk recorder, the output terminal of described target site detecting unit is connected by the high-precision fixed bit location of target site and then with the input end of driving condition identification and analysis unit, and the output terminal of described driving condition identification and analysis unit is connected with the input end of alarm modules.

3. a kind of high-precision driving behavior identification according to claim 1 and supervisory system, is characterized in that: be also included as the vehicle power of high resolving power dual camera power supply and be the power module of hard disk recorder and the power supply of industrial computer processing enter.

4. high-precision driving behavior is identified and a method for supervising, it is characterized in that: comprising:

A, by dual camera, obtain real-time video frame;

B, the real-time video frame obtaining is carried out to target site detection and target site feature extraction, thereby obtain real-time target genius loci point;

C, real-time target genius loci point and the standard target genius loci point obtaining according to training sample and To Template are compared, and according to the result of comparison, driver's driving behavior is carried out to identification and analysis;

D, by the result feedback of identification and analysis to user and show in real time.

5. a kind of high-precision driving behavior identification according to claim 4 and method for supervising, is characterized in that: described target site comprises head and people's face.

6. a kind of high-precision driving behavior according to claim 5 is identified and method for supervising, it is characterized in that: described step B, and it comprises:

B1, according to default screening conditions, the frame of video of obtaining is carried out to the detection of people's face, thereby detect facial image, described default screening conditions are:

Wherein, m is principal function, and K is objective function, and τ is visual coefficient, for area coefficient, for the scale-up factor detecting;

B2, facial image is carried out to eyes location, normalized and feature point extraction, thereby extract real-time face unique point.

7. a kind of high-precision driving behavior according to claim 6 is identified and method for supervising, it is characterized in that: in described step B2, facial image is carried out to eyes and locate this step, it is specially:

According to default unique point computing formula, facial image is carried out to eyes location, thereby orient eye areas, described default unique point computing formula is:

$E\left\{\underset{x\&RightArrow;\&infin;}{\lim }\right[L_D\left(h\right)-L_S\left(h\right)\}=E\{\underset{x\&RightArrow;\&infin;}{\lim }[L_A\left(h\right)-L_N\left(h\right)]\},$

Wherein, E is unit picture element point, and L is the region area of pixel x, and D, S, A, N are respectively four angle points of pixel x, the impact point of h for locating.

8. a kind of high-precision driving behavior according to claim 7 is identified and method for supervising, it is characterized in that: in described step B2, facial image is normalized to this step, it is specially:

According to the distance of two eyes, eye areas is normalized, thereby obtains effective human face region, described effective human face region meets:

Wherein, the radius that k is eye pupil, d is the distance between two, for pixel coefficient, for luminosity coefficient, V and D are the coordinate points between two eyes, and L is principal function.

9. a kind of high-precision driving behavior according to claim 6 is identified and method for supervising, it is characterized in that: in described step B2, facial image is normalized to this step and also comprises the step of facial image being carried out to figure image intensifying, described in carry out the formula that the step of figure image intensifying adopts and be:

$\underset{j\&Element;\left[d\right]}{\min }\underset{\mathrm{\&theta;}\&Element;\left[R\right]}{\min }(\underset{\mathrm{\&tau;},y_{\mathrm{\&tau;}}=1}{\mathrm{\&Sigma;}}{D}_{\mathrm{\&tau;}}{k}_{[{\mathrm{\&chi;}}_{\mathrm{\&tau;}},j>\mathrm{\&theta;}]}+\underset{\mathrm{\&tau;},y_{\mathrm{\&tau;}}=-1}{\mathrm{\&Sigma;}}{D}_{\mathrm{\&tau;}}{k}_{[{\mathrm{\&chi;}}_{\mathrm{\&tau;}},j\&le;\mathrm{\&theta;}]})\&le;1,$

Wherein, j arrives the span of central point d for take people's face central authorities as radius, θ be face center to the span of check point R, D is original image pixels, K is the object pixel function after processing, χ _τ, j is the pixel in processing region.

10. a kind of high-precision driving behavior according to claim 4 is identified and method for supervising, it is characterized in that: described step C, and it comprises:

C1, real-time target genius loci point and the standard target genius loci point obtaining according to training sample and To Template are compared, thereby video object image is identified, the described formula that video object image is identified to institute's foundation is:

Wherein, α is pixel coefficient, and K is definite value constant, x _τbe the horizontal ordinate of video object image, υ _jfor video object image, can follow the tracks of the ordinate of coefficient, γ is gamma, and i, j are ordered series of numbers coefficient;

C2, according to identification result driver's driving behavior is analyzed.