CN109583535B - Vision-based logistics barcode detection method and readable storage medium - Google Patents

Abstract

The invention belongs to the technical field of logistics detection, and discloses a visual-based logistics barcode detection method and a computer program; including image capture; identifying and locating cargo positions; identifying and locating barcode positions; reading barcodes; and error correction and alarming. The invention uses a computer vision library, adds noise removal means, locates the position and rotation angle of the one-dimensional code in a complex background, and intercepts the horizontal one-dimensional code picture; ensures that each product only corresponds to one label, and adds the found goods The mechanism of alarm when there is no label; avoid the calculation of irrelevant pixel data in the picture, and improve the calculation speed. The invention realizes the delivery of parcels on the assembly line, installs a high-definition camera at a fixed position, identifies the location of the logistics parcel within the field of view of the camera, obtains and decodes the one-dimensional code information pasted on the goods, and then completely replaces the manual scanning of the traditional assembly line.

Description

A vision-based logistics barcode detection method and readable storage medium

Technical Field

The present invention belongs to the technical field of logistics detection, and in particular relates to a vision-based logistics barcode detection method and a readable storage medium.

Background Art

At present, the commonly used existing technologies in the industry are as follows:

With the widespread popularization of the Internet, the logistics industry has also developed rapidly. The pursuit of improving logistics speed, increasing the proportion of information and intelligence in logistics, and replacing part of the manpower have become an inevitable development trend in the logistics industry. In the assembly line operation of the logistics industry, the use of one-dimensional barcodes to number and mark logistics packages has become a common practice in the industry, and most logistics companies still need to manually scan the barcodes of packages with handheld barcode scanners. This practice not only increases labor costs, but also because the scanning direction of the scanner needs to be horizontally aligned with the one-dimensional barcode and the scanning distance is limited, the operator must find and align the barcode position and angle manually. When the logistics volume increases, this method will be difficult to meet the speed requirements.

At present, with the widespread popularity of mobile camera equipment, it is becoming easier and easier to obtain digital images. At the same time, artificial intelligence technology has developed rapidly this year, and visual application technology has been continuously developed. There are sufficient basic conditions for developing industrialized one-dimensional barcode information reading algorithms based on visual recognition technology.

There are already one-dimensional barcode recognition algorithms in the prior art, and some mobile phone APP applications can also recognize one-dimensional barcodes under the premise of aligning with horizontal one-dimensional codes. However, the one-dimensional codes on logistics goods are pasted at random and non-horizontally, and the condition that only a single barcode is recognized for a single item must be met. Simply recognizing the barcode does not play a role in verifying the correspondence. At present, there is still a lack of methods to determine whether a single item corresponds to a unique barcode number. The integrity and practicality of the existing visual recognition barcode function do not meet the requirements of the logistics industry.

In summary, the problems existing in the prior art are:

(1) Existing logistics also needs to meet the condition that only a unique barcode can be identified for a single item. If a barcode label is missing for an item, that is, the item on the assembly line cannot be matched with a unique barcode, and the system does not detect this situation and issue a warning, it will eventually cause a package without information to enter the sorting assembly line. If it is possible to predict and alarm, the situation of missing labels can be reported in time and the operator can be reminded to handle it.

(2) The barcodes on existing logistics goods are pasted at random and non-horizontal angles, and current recognition programs such as ZBar require the barcode to be in a horizontal position for correct recognition. When the barcode has a large horizontal angle deviation in the image, it will lead to the inability to read the correct barcode code. In addition, multiple barcodes will appear in the camera's field of view at the same time. When multiple barcodes coexist, it is difficult to find a fixed rotation direction to turn all barcodes back to a horizontal angle for reading. Therefore, this technology must rotate all barcodes in the field of view to a readable angle and read them to ensure that the barcode number can be read correctly and no barcodes are missed.

The difficulty and significance of solving the above technical problems:

(1) How to determine how many goods and barcodes are in the camera's field of view, and under what circumstances can it be determined that there are no barcode labels affixed to the goods. Solving this problem can detect situations where there are no barcode labels affixed to the goods.

(2) How to decode the numbers represented by the barcode label attached to the goods, and the length of the numbers decoded from the barcode label must conform to the actual barcode number format. Solving this problem can prevent misreading of barcodes that do not conform to the correct encoding method.

(3) A certain fault tolerance mechanism needs to be designed. For example, noise interference may cause errors in the number and accuracy of identifying boxes or reading labels in a certain frame or several frames of images, but no false alarms will be issued. Due to image shooting noise interference and other reasons, the program cannot guarantee the accuracy and resolution of objects in each frame of the picture, and the accuracy of barcode reading can reach 100%. Solving this problem can prevent sensitivity errors.

Summary of the invention

In view of the problems existing in the prior art, the present invention provides a vision-based logistics barcode detection method and a readable storage medium.

The present invention is implemented as follows: a visual-based logistics barcode detection method specifically includes the following steps:

Step 1: Image capture: Fix the high-definition high-speed camera at a certain height above the conveyor belt, ensure that the field of view includes a certain length of the conveyor belt and that clear images of the cargo outline and labels can be captured under the premise of cargo movement. After setting the camera mode, the camera starts to take high-definition images, and the program converts the images into cv::Mat format;

Step 2: Identify and locate the cargo: pre-arrange a YOLO target detection model, calculate the image in step 1, and obtain all the object types (cargo, labels) in the captured image and their coordinate positions (xy coordinate values of the object in the image), bounding box size (length of the image pixels occupied); in this step, save the detected objects into different data types, in which the coordinate position of each object, the bounding box size, and the image captured by the bounding box in the original image are saved;

Step 3: Identify and locate the barcode position: After the various objects are identified and saved in step 2, the program first processes the "label" type of objects. The intercepted images saved by the "label" type of objects are screened and corrected by the barcode processing program to obtain the horizontal barcode image.

Step 4: Read the barcode: Use the Zbar method to identify the barcode. After the above steps, a horizontal barcode area image is obtained. Use the Zbar program to identify the image and get the barcode number.

Step 5: Error correction alarm: Match the "goods" objects identified in the camera's field of view with the numbers decoded from the "label" images. If a "goods" object is found to be missing the corresponding label number in multiple consecutive frames of images, an alarm will be issued and the operator will be notified to handle it.

Furthermore, in step one, the shooting mode of the high-definition high-speed camera is set. A grating can be installed in front of the camera's field of view, and the shooting mode can be set to a trigger mode. When goods are about to enter the shooting range, the grating is triggered, commanding the camera to take pictures continuously for a period of time; the shooting mode can also be set to continuous photography, regardless of whether there are goods on the conveyor belt, continuous photography is taken at a certain time interval.

Furthermore, in step 2, the types of objects to be detected and identified are divided into "goods" (hexahedral boxes or bagged packages) and "labels" (barcode labels), and the center position (coordinates) and bounding box size (width and height in the image coordinate system) of the object in the photo are located; the method used is the YOLO target detection algorithm in deep learning, and its computing speed can achieve the effect of real-time detection, which meets the target requirements; the specific steps are as follows:

(1) The present invention pre-trains a reliable object detection model and deploys the model algorithm in the program. The model is a mature model formed by collecting a large amount of actual image data and performing training operations. When the camera takes a picture, the model will perform target detection on the goods in the picture.

(2) The object detection model will process the captured images one by one to obtain the type, number, and bounding box information of the objects in the captured images. Since multiple objects are expected to appear in the same field of view, the detection model needs to detect all "goods" and "label" objects and design a specific data format for them, in which the object type (box, bagged package, label), coordinate position (xy coordinate value of the object in the image), bounding box size (length of the image pixels occupied) and other information are saved. At the same time, the rectangular boxes they occupy are cut from the original image into sub-images and saved in the data format, which are sent to the next step of the one-dimensional barcode detection program for calculation. If there are no goods in the field of view, the image is not cut and this step is continued. The designed data format is as follows.

Cargo type:

The name variable records whether it is a box or a package, the center_position variable records the object coordinates, the bounding variable records the range of the rectangular box, the area matrix stores the rectangular box sub-image captured from the original image, and the link_code variable indicates whether the small block corresponds to a barcode number.

Tag class:

The center_position variable records the label coordinates, area stores the rectangular sub-image captured from the original image, num_type represents the decoded barcode type, num_code represents the digital content, and link_packegebox_num represents the number of the "goods" object to which the label is attached.

Furthermore, in step 3, since the one-dimensional code image is easy to be read by the code reading program only when it is in the horizontal direction and the pixels are clear and the irrelevant pixels are few, the barcode recognition algorithm in the present invention utilizes the characteristics of the one-dimensional barcode image with obvious outline and consistent image gradient, and searches for the area with consistent gradient change direction in the intercepted image of the identified "label" type object as the area where the one-dimensional code may exist. The specific implementation process is:

(1) The image P0 of the “label” object captured in step 2 (a sub-image captured from the original image) is converted into a grayscale image by grayscale processing, and the noise in the image is removed by Gaussian blur to obtain a grayscale image P1;

(2) Use the Cany operator to calculate the gradient values of each pixel in the image P1 in two directions, that is, two image matrices Gx and Gy. Add the image matrices Gx and Gy to obtain an image matrix P2 with obvious visual contour features. In P2, the contour of the barcode can be clearly identified. The main gradient value and direction of each contour point can be calculated based on the gradient values in the XY directions. The calculation formula is:

d＝sqrt(x(I)^2+y(I)^2)

(3) Use a rectangle of size m*m to divide the image P2 into M small blocks. The data structure of each small block is:

This data structure class can save the pixel range of the small block in the image, the number of points in all contour images P2 that fall into the area of the small block, and the main gradient direction of the points.

(4) Perform screening to remove small blocks of images that do not contain contour points or contain too few contour points, and only retain small blocks with more than m/2 contour points (anchor_block);

(5) Calculate the main gradient direction of the remaining image blocks. Starting from 0 degrees, each 20 degrees is divided into a category, with a total of 9 angle ranges from 1 to 9, and the category number is determined according to the θ value of the gradient direction of each contour point, that is, the belong_angle value in the anchor_block data structure class. When the number of contour points in a certain angle range accounts for more than 60% of the total number of contour points, the angle range is determined as the main angle direction of this small block. If not, it means that the gradient in this small block is relatively chaotic, and the small block is deleted.

(6) Connect the small blocks with the same gradient direction. Cluster the small blocks with pixel distance within d and the same main gradient direction into one region, and use the gradient direction as the gradient direction of the merged "region". In this step, small blocks that are far away and too few to be clustered will be eliminated.

(7) Determine whether the area of all merged regions is above the threshold S. If so, keep it; otherwise, remove it.

(8) After the above steps, it can be concluded that the remaining area is the location of the barcode, and the gradient direction θ of the area obtained in (6) is the horizontal angle deviation of the barcode. After rotating θ degrees clockwise, a recognizable one-dimensional barcode image is obtained.

Furthermore, in step 4, the recognition method is an open source program that can easily read the information of the horizontal one-dimensional barcode, and the output result is a digital string of a certain length. The barcode is generally one of the UPC, EANEAN, Code39, and Code128 codes. If the recognition is successful, the type of the target barcode will be output together.

Further, in step 5, the logic design of this step is as follows: after steps 1 to 4, if the corresponding coordinate axis position of each "label type" object that can be correctly read and coded is located within the rectangular frame of a "goods" type object, it is determined that the "goods" type object has been marked and coded. When it is found that there are "goods" objects that are not marked, it means that the detection model has not detected the corresponding "label type" object, and the program records the number of non-correspondences. If there are multiple consecutive frames of pictures that do not correspond, the number of non-correspondences will continue to increase. If it is higher than a certain value, an alarm will be issued to remind the operator that the barcode has not been posted on the goods. If all the "goods" type objects detected in a certain frame of the picture can correspond to a number, the number of non-correspondences will be cleared. At the same time, considering that when the object just enters the field of view, the "goods" has been identified but the label has not entered the field of view, when the "goods" type object is at the boundary of the field of view and does not correspond to the code, the number of non-correspondences will not increase. The above process can avoid the occurrence of sensitive alarms caused by interference from individual uncertain factors. An alarm will only occur when it is determined that there is an object whose label number cannot be read in multiple frames of images.

The barcode selected for reading in logistics is generally one of the following codes: UPCUPC, EANEAN, Code39, and Code128. If the digital type of the barcode detected in step 4 is not the selected type, this number cannot be used to correspond to the "goods" object.

In summary, the advantages and positive effects of the present invention are:

The source code of the cargo detection network structure of the present invention is written in C/C++ language. The prototype is the YOLO target detection network structure. In the present invention, the input layer and output layer structures are changed, and data is collected by itself to train the network model, and a stable algorithm structure has been formed; the source code, network configuration and network parameters in the present invention are protected. After testing 853 cargoes and labels contained in 543 pictures, 827 object types can be correctly identified, the classification recognition accuracy rate reaches 96.96%, and the area measurement accuracy rate reaches 90.42%;

The present invention uses a method for finding a one-dimensional barcode by using a gradient. The source code is independently written in C++ language. The principle is to detect contour points and the gradient directions of the contour points, thereby screening out the area with the most consistent gradient reverse direction as the area where the one-dimensional code is located. The computer vision library (OPENCV) is used, and noise removal means are added to locate the position of the one-dimensional code and its rotation angle in a complex background, and a horizontal one-dimensional code image can be captured.

The goods and labels comparison inspection method and source code of the present invention can ensure that each goods corresponds to only one label, and add an alarm mechanism when the goods are found to be without labels. At the same time, it takes into account that when the goods are not completely in the field of view, the barcode may not enter the field of view and the reading may be incorrect, so as to avoid false alarms. In the test on the assembly line, 38 goods without labels were mixed in 319 goods. Except for 2 goods that passed without alarm, the rest were able to detect the unpasted barcode information and alarm.

The method of combining deep learning with geometric image processing in the present invention uses deep learning to extract the segmented image of each cargo, and then uses geometric methods to process the segmented image, avoiding the calculation of irrelevant pixel data in the image, improving the calculation speed, and realizing real-time detection. The average time for testing single-image target recognition is 0.04s, the average time for barcode recognition is 0.05, the average time for calculating each frame of the image is 0.09s, and the FPS is 11.

The present invention realizes that while the package is being transported on the assembly line, a high-definition camera is installed at a fixed height above the assembly line to identify the location of the logistics package within the camera's field of view, locate the one-dimensional code image attached to the goods and decode and read it, thereby completely replacing the traditional manual scanning of the assembly line; it has the following advantages:

(1) The recognition speed accuracy can meet the requirements of the express delivery industry, and the barcode is basically not missed within the conveyor belt speed range;

(2) Within the camera’s field of view, it can recognize and only recognize the one-dimensional barcode label affixed to the logistics package (usually located on the surface of the express packaging bag or express box). If it is found that the logistics package does not have a barcode affixed, it will actively alarm;

(3) It can accurately locate the position of the one-dimensional code label. If the one-dimensional code has a horizontal deviation angle, the angle can be solved through an algorithm and the code can be automatically rotated to align.

(4) It can read and decode one-dimensional code information and convert it into digital numbers. It can also check whether the digital format is correct and discard invalid numbers read out.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a vision-based logistics barcode detection method provided in an embodiment of the present invention.

FIG. 2 is a technical roadmap of a vision-based logistics barcode detection method provided in an embodiment of the present invention.

FIG. 3 is a flow chart of cargo detection provided by an embodiment of the present invention.

FIG. 4 is a flow chart of barcode detection and reading provided by an embodiment of the present invention.

DETAILED DESCRIPTION

In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

The application principle of the present invention is further described in detail below with reference to the accompanying drawings;

As shown in FIG1 , the visual-based logistics barcode detection method provided by an embodiment of the present invention specifically includes the following steps:

S101: Image capture: Fix a high-definition high-speed camera at a certain height directly above the conveyor belt, ensuring that a certain length of the conveyor belt is included in the field of view and that clear cargo outlines and label images can be captured while the cargo is moving;

S102: Using the YOLO target detection algorithm in deep learning, determine the types of goods, label positions, and bounding box sizes of all goods in the captured images; and save the data types of various goods and labels;

S103: Processing the captured image saved in the label data in step S102, reducing noise and rotating it to a horizontal position, obtaining an easily readable one-dimensional barcode image, and making the goods data correspond to one label data in space;

S104: Read the barcode: Use the Zbar method to identify the barcode. After the above steps, a horizontal image of the one-dimensional code image is obtained, and the barcode data is obtained by using the Zbar program for identification.

S105: Error correction alarm: When an error occurs, an alarm will be prompted and manual correction will be performed if necessary.

In step S101, the high-definition high-speed camera provided by the embodiment of the present invention can install a grating in front of the camera's field of view, and the shooting mode can be set to a trigger mode. When goods are about to enter the shooting range, the grating is triggered, and the camera is commanded to take pictures continuously for a period of time; the shooting mode can also be set to continuous photography, regardless of whether there are goods on the conveyor belt, continuous photography is performed at a certain time interval.

In step S102, the types of logistics goods provided by the embodiment of the present invention are divided into hexahedral boxes and bagged packages, and the center position (coordinates) of the goods in the photo and the size of the bounding box (width and height in the image coordinate system) are located; at the same time, the center position of the label and the size of the bounding box are also identified, and the sub-image of the label and the goods is cut out for the next analysis; the method used is the YOLO target detection algorithm in deep learning, and its computing speed can achieve the effect of real-time detection, which meets the target requirements; the specific steps are as follows:

(1) The present invention will pre-train a reliable object detection model and deploy the model algorithm in the program; when the camera takes a picture, the program will perform target detection on the goods and labels in the picture;

(2) The object detection model will process the captured images one by one to obtain the center position of the goods and labels in the captured images and their bounding box information; since multiple objects are expected to appear in the same field of view, the detection model needs to detect all objects and cut them into sub-images in turn, which will be sent to the next step of the one-dimensional barcode detection program for calculation; if no labels or goods appear in the field of view, no image cutting will be performed and this step will continue.

In step S103, the one-dimensional code image provided by the embodiment of the present invention is easy to be read by the code reading program only when it is in the horizontal direction and the pixels are clear. The barcode recognition algorithm uses the characteristics of the one-dimensional barcode image with obvious contour and consistent image gradient, and judges that the pixel areas in the image that meet the characteristics and are concentrated in the area are one-dimensional codes, and the gradient directions of these areas are used as the rotation angles of the one-dimensional codes. The specific implementation process is:

(1) Grayscale each label sub-image and perform Gaussian blur and noise reduction processing;

(2) Use the Cany operator to calculate the contour points and gradient directions of the contour points in the image;

(3) Use a rectangle of size m*m (the value of m is adjusted according to the actual effect, generally 5-10) to divide the cargo sub-image into several small blocks, and record the contour points and gradient directions of the contour points contained in each small block;

(4) Perform screening to remove small image blocks that do not contain contour points or contain too few contour points;

(5) Calculate the main gradient direction of the remaining image blocks, that is, take the direction in which the gradient directions of the contour points in the block are most concentrated as the main gradient direction; if the gradient direction in the block is relatively scattered and chaotic, remove it;

(6) Connect the small blocks with the same gradient direction, cluster and merge the small blocks with similar distance and concentrated density into one area, and use the gradient direction as the gradient direction of the merged area; remove the small blocks with large distance and small number that cannot be clustered;

(7) Determine whether the area of the merged region is above a certain threshold. If so, keep it; otherwise, remove it;

(8) After the above steps, it can be concluded that the remaining area is the location of the one-dimensional code, and the horizontal angle deviation of the one-dimensional barcode can be obtained according to the gradient direction. After rotation, a recognizable one-dimensional barcode can be obtained.

In step S104, the recognition method provided by the embodiment of the present invention is an open source program, which can easily read the information of the horizontal one-dimensional barcode, and its output result is a digital string of a certain length.

In step S105, the embodiment of the present invention provides an alarm when the following error occurs, and manual correction is performed when necessary:

(1) The barcode is not pasted, pasted in the blind spot of the camera's field of view, or the barcode is severely damaged and cannot be read; when it is found that the goods have completely entered the field of view and no readable barcode is found after processing from step 1 to step 4, an alarm is issued;

(2) Due to label printing problems, more than two barcode areas may be detected in step 4; if it is found that the decoded numbers of the two barcodes are inconsistent, or the decoded numbers have lengths that do not meet the standard, an alarm is issued;

(3) Since the photo-taking time is short, the same goods may appear in multiple consecutive photos. The present invention will estimate a permissible length of time for the repeated appearance of the same serial number based on the conveyor belt speed. After reading a new barcode, the barcode with the same serial number is allowed to be read repeatedly within a certain period of time thereafter. If the time is exceeded, it indicates that the conveyor belt has not conveyed the goods forward and an alarm is issued.

As shown in FIG2 , a technical roadmap of a vision-based logistics barcode detection method is provided in an embodiment of the present invention.

As shown in FIG3 , a flow chart of cargo detection provided by an embodiment of the present invention is shown.

As shown in FIG. 4 , a barcode detection and reading flow chart is provided in an embodiment of the present invention.

The application principle of the present invention is further described in detail below in conjunction with specific embodiments;

The present invention first intercepts the boundary box of the goods within the field of view to form a sub-image of each product, then processes and merges the contours and gradient directions in these sub-images, and finally extracts the pixel range belonging to the one-dimensional bar. After the finally extracted one-dimensional code image is rotated to horizontal, the one-dimensional code sequence is read out using a decoding program.

The specific process of the present invention is as follows:

(1) Start the processing flow and set the camera mode to ensure that every product and its label image can be clearly captured;

(2) Identify the location and rectangular size of each item and label, and extract all item and label sub-images;

(3) Find the position of the one-dimensional code in each label sub-image and extract the one-dimensional code image;

(4) Read the digital code of the one-dimensional code;

(5) If an error occurs, an alarm is issued; otherwise, the process continues from (2) to the next image.

In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When the use is implemented in whole or in part in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL) or wireless (e.g., infrared, wireless, microwave, etc.) mode) to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims (5)

1. A visual-based logistics barcode detection method, characterized in that the visual-based logistics barcode detection method specifically includes the following steps: Step 1: Fix the high-definition high-speed camera at a certain height directly above the conveyor belt, ensuring that a certain length of the conveyor belt is included in the field of view and that clear cargo outlines and label images can be captured while the cargo is moving; Step 2: Use the YOLO target detection algorithm in deep learning to determine the types of goods, label positions, and bounding box sizes of all goods in the captured images; save the data types of various goods and labels; Step 3: Process the captured image saved in the tag data in step 2, reduce noise and rotate it to a horizontal position, and obtain an easy-to-read one-dimensional barcode image, so that the goods data can correspond to one tag data in space; Step 4: Use the Zbar method to identify the barcode. After the above steps, a horizontal image of the one-dimensional code image is obtained. Use the Zbar program to identify and obtain the barcode data. Step 5: When an error occurs, an alarm will be prompted and manual correction will be performed if necessary; In the step 1, a high-definition high-speed camera can be equipped with a grating in front of the camera's field of view, and the shooting mode can be set to a trigger mode. When goods are about to enter the shooting range, the grating is triggered, and the camera is commanded to take pictures continuously for a period of time; the shooting mode can also be set to continuous shooting, regardless of whether there are goods on the conveyor belt, continuous shooting is performed at a certain time interval; In the step 2, the types of logistics goods are divided into hexahedral boxes and bagged packages, and the center position and bounding box size of the goods in the photo, the center position and bounding box size of the label in the photo are located; since the label only exists on the surface of the goods, the sub-image of the label will be intercepted and saved into the label data for the next step of analysis; YOLO target detection algorithm in deep learning; The specific steps are as follows: (1) Pre-train a reliable object detection model and deploy the model algorithm in the program; when the camera takes a picture, the program will perform object detection on the goods in the picture; (2) The object detection model will process the captured images one by one to obtain the number, location and bounding box information of the goods and labels in the captured images; since multiple objects are expected to appear in the same field of view, the detection model needs to detect all objects and cut them into sub-images in turn, save them as corresponding class data and send them to the next step of the one-dimensional barcode detection program for calculation; if there are no goods or labels in the field of view, no image cutting is performed and this step continues; In the step 3, the one-dimensional code image is easy to be read by the code reading program only when it is in the horizontal direction and the pixels are clear. The barcode recognition algorithm uses the characteristics of the one-dimensional barcode image with obvious outline and consistent image gradient, and judges the pixel area in the image that meets the characteristics and is concentrated in the area as a one-dimensional code, and uses the gradient direction of the area as the rotation angle of the one-dimensional code; the specific implementation process is: (1) Grayscale the sub-image of the label data, and perform Gaussian blur and noise reduction processing; (2) Use the Cany operator to calculate the contour points and gradient directions of the contour points in the image; (3) Use an m*m rectangle to divide the cargo sub-image into several small blocks, and record the contour points and gradient directions of the contour points contained in each small block; (4) Perform screening to remove small image blocks that do not contain contour points or contain too few contour points; (5) Calculate the main gradient direction of the remaining image blocks, that is, take the direction in which the gradient directions of the contour points in the block are most concentrated as the main gradient direction; if the gradient direction in the block is relatively scattered and chaotic, remove it; (6) Connect the small blocks with the same gradient direction, cluster and merge the small blocks with similar distance and concentrated density into one area, and use the gradient direction as the gradient direction of the merged area; remove the small blocks with large distance and small number that cannot be clustered; (7) Determine whether the area of the merged region is above a certain threshold. If so, keep it; otherwise, remove it; (8) After the above steps, it can be concluded that the remaining area is the location of the one-dimensional code, and the horizontal angle deviation of the one-dimensional barcode can be obtained according to the gradient direction. After rotation, a recognizable one-dimensional barcode can be obtained. 2. The vision-based logistics barcode detection method as described in claim 1 is characterized in that in step 4, the recognition method is an open source program that can easily read the information of the horizontal one-dimensional barcode, and its output result is a digital string of a certain length. 3. The visual-based logistics barcode detection method according to claim 1, characterized in that in step 5, an alarm will be prompted when the following errors occur, and manual correction will be performed if necessary: (1) The barcode is not pasted, pasted in the blind spot of the camera's field of view, or the barcode is severely damaged and cannot be read; When it is found that the goods have completely entered the field of view, and no readable barcode is found after processing from step 1 to step 4, an alarm is issued; (2) Due to label printing problems, more than two barcode areas may be detected in step 4; if it is found that the decoded numbers of the two barcodes are inconsistent, or the decoded numbers have lengths that do not meet the standard, an alarm is issued; (3) Since the photo-taking time is short, the same goods may appear in multiple consecutive photos. The present invention will estimate a permissible length of time for the repeated appearance of the same serial number based on the conveyor belt speed. After reading a new barcode, the barcode with the same serial number is allowed to be read repeatedly within a certain period of time thereafter. If the time is exceeded, it indicates that the conveyor belt has not conveyed the goods forward and an alarm is issued. 4. An information data processing terminal for implementing the vision-based logistics barcode detection method described in any one of claims 1 to 3. 5. A computer-readable storage medium, comprising instructions, which, when executed on a computer, enables the computer to execute the vision-based logistics barcode detection method as described in any one of claims 1 to 3.

Images