Container marking identification method and device, computer device and storage medium

Abstract

The application relates to a target mark identification method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a video stream obtained by monitoring a target; performing image detection processing on a video frame in the video stream, screening a video frame comprising the target from the video frame as a target video frame; performing target tracking processing based on the target video frame to obtain a driving track of the target; when it is determined based on the driving track that a current video frame in the target video frame meets a preset target condition, performing mark identification based on the current video frame to obtain corresponding target mark information. The method can improve the mark identification accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, in particular to a target mark recognition method and device, computer equipment and a storage medium. BACKGROUND

[0002] Port is an important cargo collection point of logistics transportation, which plays an important role in international logistics and domestic logistics. Containers are the most common cargo storage and packaging method, which has a wide application in port cargo management and transportation. There are many container trucks entering and leaving the port every year or every day, so the registration and management of containers entering and leaving the port are essential for port logistics, cargo management and safety management.

[0003] Each container has its corresponding mark (located on one side of the end door of the container), through which the source and other information of the container can be identified. Artificial observation for container registration is time-consuming and labor-intensive, and the cost is also large. At present, the traditional way is to intelligently identify and register the container mark automatically through a camera, for example, a traditional computer vision method is used to determine the position of the container, and then some common morphological transformations are performed, such as binarization, dilation and erosion, and then statistical or template character matching principles are used to extract the container mark. However, the traditional computer vision method needs manual feature design, and the stability is poor for complex scenes, and there is a problem of inaccurate recognition. SUMMARY

[0004] Therefore, it is necessary to provide a target mark recognition method and device, computer equipment and a storage medium capable of improving the accuracy of mark recognition in view of the above technical problems.

[0005] A container mark recognition method, the method comprising:

[0006] obtaining a video stream obtained by monitoring a target;

[0007] performing image detection processing on a video frame in the video stream, and screening a video frame including a target from the video frame as a target video frame;

[0008] performing target tracking processing based on the target video frame to obtain a driving trajectory of the target;

[0009] when it is determined based on the driving trajectory that a current video frame in the target video frame meets a preset target condition, performing mark recognition based on the current video frame to obtain corresponding target mark information.

[0010] A container mark recognition device, the device comprising:

[0011] acquire a video stream obtained by monitoring a target;

[0012] screen a video frame in the video stream through image detection processing, and screen out a video frame including the target from the video frame as a target video frame;

[0013] perform target tracking processing based on the target video frame, and obtain a driving track of the target;

[0014] when it is determined that a current video frame in the target video frame meets a preset target condition based on the driving track, perform mark identification based on the current video frame, and obtain corresponding target mark information.

[0015] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0016] acquire a video stream obtained by monitoring a target;

[0017] screen a video frame in the video stream through image detection processing, and screen out a video frame including the target from the video frame as a target video frame;

[0018] perform target tracking processing based on the target video frame, and obtain a driving track of the target;

[0019] when it is determined that a current video frame in the target video frame meets a preset target condition based on the driving track, perform mark identification based on the current video frame, and obtain corresponding target mark information.

[0020] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the following steps:

[0021] acquire a video stream obtained by monitoring a target;

[0022] screen a video frame in the video stream through image detection processing, and screen out a video frame including the target from the video frame as a target video frame;

[0023] perform target tracking processing based on the target video frame, and obtain a driving track of the target;

[0024] when it is determined that a current video frame in the target video frame meets a preset target condition based on the driving track, perform mark identification based on the current video frame, and obtain corresponding target mark information.

[0025] The target mark identification method, device, computer device and storage medium described above, through image detection processing on the video stream obtained by monitoring the target, filter out the target video frame including the target. Further, based on the filtered target video frame, the accurate tracking of the target is realized. Based on the tracking result, a clear image frame containing the complete target can be accurately and reliably extracted from the video stream, the text recognition is performed and the target mark information is extracted. The present application does not need manual design of features, and can automatically capture a video frame meeting the preset target condition in the monitoring video stream scene, so as to accurately identify the target mark based on the video frame. And the whole process is automatic detection and identification, which greatly improves the target mark identification efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 An application environment diagram of the target mark identification method in one embodiment;

[0027] Figure 2 A flowchart of the target mark identification method in one embodiment;

[0028] Figure 3 A flowchart of the step of image detection processing on the video frame in the video stream in one embodiment, filtering out the video frame including the target as the target video frame;

[0029] Figure 4 A comparison diagram of the video frame containing the incomplete container and the video frame containing the complete container in one embodiment;

[0030] Figure 5 A principle diagram of the three-point snapshot mode in one embodiment;

[0031] Figure 6 A flowchart of the step of container mark information extraction based on the current video frame in one embodiment;

[0032] Figure 7 A diagram of the possible position of the box type code in the container mark in one embodiment;

[0033] Figure 8 A flowchart of the target mark identification method in one embodiment;

[0034] Figure 9 A structure block diagram of the target mark identification device in one embodiment;

[0035] Figure 10 An internal structure diagram of the computer device in one embodiment. DETAILED DESCRIPTION

[0036] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0037] The target mark identification method provided by the present application can be applied to an application environment as shown in Figure 1 . The monitoring device 102 communicates with the computer device 104 through a network. The monitoring device 102 is arranged at a preset position, such as a road checkpoint, for real-time monitoring of a target passing through the preset position. The monitoring device 102 transmits the collected video stream to the computer device 104. The computer device 104 performs image detection processing on the video frames in the video stream, and screens out the video frames including the target from the video frames as target video frames. The computer device 104 performs target tracking processing based on the target video frames to obtain a driving trajectory of the target; when it is determined based on the driving trajectory that a current video frame in the target video frames meets a preset target condition, mark identification is performed based on the current video frame to obtain corresponding target mark information.

[0038] The monitoring device 102 can be a monitoring camera. The computer device 104 can be a terminal or a server. The terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices, and the server can be implemented by an independent server or a server cluster composed of multiple servers.

[0039] In one embodiment, as shown in Figure 2 , a container mark identification method is provided. Taking the computer device in Figure 1 as an example, the method includes the following steps:

[0040] Step S202, obtaining a video stream obtained by monitoring a target.

[0041] Specifically, a monitoring device can be prearranged at a position where the target passes through. For example, when the target is a transport vehicle, a monitoring device can be prearranged at a road checkpoint for goods transportation, and the monitoring device can monitor the vehicles passing through the road checkpoint in real time. The computer device can obtain the video stream collected and transmitted by the monitoring device.

[0042] It should be noted that one of the application scenarios of the present application is to realize the transportation of goods by means of vehicles carrying containers in port logistics. When the vehicle passes through the road checkpoint, the containers entering and leaving the checkpoint usually need to be registered and managed. Therefore, a monitoring device needs to be arranged at the road checkpoint to monitor and manage the vehicles entering and leaving the road checkpoint.

[0043] In step S204, image detection processing is performed on the video frames in the video stream, and video frames including the target are filtered from the video frames as target video frames.

[0044] Specifically, the computer device can perform image detection processing on each video frame in the video stream respectively to obtain image detection results. The image detection result can be whether the target is detected in the video frame. Further, the computer device can filter video frames including the target from the video frames as target video frames based on the image detection results.

[0045] In one of the embodiments, the target can be a transport vehicle. Accordingly, the computer device can filter video frames including the transport vehicle from the video frames as target video frames. In one of the embodiments, the target can be a container vehicle, and the container vehicle is loaded with a container. When the container vehicle is detected, the computer device further judges whether the container vehicle includes a complete container. Further, the computer device can filter video frames including the container vehicle and the complete container from the video frames as target video frames.

[0046] In one of the embodiments, the computer device can send the video stream frame by frame to the algorithm input interface, and first use the target detection algorithm to detect vehicles. Because the flow vehicles are in the card hole, the detected vehicles do not necessarily include complete containers. Therefore, the detected vehicle image patch needs to be further judged for container integrity. In another embodiment, part of the frames of the video stream can be sent to the algorithm input interface, for example, every few frames or part of the frames are randomly selected from the video stream to be sent to the algorithm input interface.

[0047] In one of the embodiments, the computer device can transmit the video frames to a pre-trained image detection model, and process the video frames through the image detection model to output the target bounding box and its related attributes in each frame, and whether the bounding box contains a complete target. The related attributes of the bounding box can include the coordinates, length and width of the bounding box, and bounding box confidence, etc. The bounding box confidence represents the confidence degree of the bounding box selecting the target vehicle.

[0048] In an embodiment, the image detection model can be trained by taking images including complete targets as positive samples, and taking images not including targets or only including incomplete targets as negative samples. In an embodiment where the target is a container vehicle, the image detection model can be specifically trained by taking images of vehicles with complete containers as positive samples, and taking images of vehicles with incomplete containers as negative samples. The sample region of the vehicle is marked in the positive samples, and the complete container is included; the sample region of the vehicle is marked in the negative samples, and the incomplete container is included. Of course, images not including vehicles can also be taken as negative samples for training. In this way, through training of positive samples and negative samples, the image detection model can recognize vehicles and can recognize whether complete containers are included on the vehicles.

[0049] In step S206, target tracking processing is performed based on the target video frames to obtain a driving trajectory of the target.

[0050] The target tracking processing refers to tracking and marking the same target appearing in different frames, and in this application, it mainly ensures that the target appearing in consecutive frames is the same object, so that the motion route of the target can be found.

[0051] In an embodiment, step S206, i.e., performing target tracking processing based on the target video frames to obtain a driving trajectory of the target, includes: obtaining position information of a bounding box for selecting a target in the target video frames; performing target tracking processing based on the position information of the bounding box to obtain object identifiers respectively corresponding to the bounding boxes; and determining the driving trajectory of the target based on position information of the bounding boxes corresponding to the same object identifier in each target video frame.

[0052] In some embodiments, the computer device can input the target video frames including the bounding box of the target into the tracking algorithm, and the video frames determined as not including the target or not including a complete target are not input into the tracking algorithm. The bounding box including the target can be specifically a bounding box for selecting the target obtained by performing target detection on the video frames.

[0053] In some embodiments, when the target is a container vehicle, the computer device can input the target video frames including the bounding box of the complete container into the tracking algorithm, and the video frames determined as including an incomplete container are not input into the tracking algorithm. The bounding box including the complete container can be specifically a bounding box for selecting and marking the target vehicle obtained by performing target detection on the video frames.

[0054] Further, the computer device performs target tracking processing on the input target video frame, and returns an object identifier (also referred to as a track id) of the target in the target video. If the tracked target in different target video frames is the same target, the same track id is returned. In this embodiment, the computer device can use a sort (Simple Online And Realtime Tracking) tracking algorithm based on Kalman filtering, and of course other tracking algorithms can also be used, such as a multi-target tracking algorithm based on a multi-thread single-target tracker, or a multi-target tracking algorithm based on deep learning end-to-end, etc., which are not limited in the embodiments of the present application.

[0055] Further, the computer device can determine that the bounding boxes corresponding to the same object identifier in each target video frame are the same target, and further determine the driving trajectory of the target based on the position information of the bounding boxes corresponding to the same object identifier.

[0056] In some embodiments, in order to prevent the loss of tracked targets (i.e., marking the same target as different object identifiers), the present application can not track the frame in which the target is lost. The specific method is as follows: for the current detection frame currently processed in the video stream, if a target is detected, it can be sent to the tracking algorithm, and the tracking algorithm updates the current tracking target (such as a container vehicle). If the object identifier of the tracking target of the frame is the same as that of the previous frame, it means that it is the same target, otherwise if the detection result is empty, the tracking target is not updated, which can prevent the loss of tracking targets.

[0057] In addition, if the current tracked object identifier is not detected in a continuous preset frame number (T miss frames), the tracked target object can be discarded, which means that the target has left the camera area, and the object appearing thereafter is considered to be a newly appearing target using a new object identifier. In the embodiments, the preset frame number can be the frame number corresponding to the original video frame, for example, T miss is set to 100.

[0058] In one of the embodiments, the step of performing target tracking processing based on the position information of the bounding box to obtain the object identifier corresponding to each bounding box specifically includes: for the current tracking frame subjected to target tracking processing, when the object identifier determined based on the position information of the bounding box in the current tracking frame is the same as the object identifier of the previous frame, the object identifier is taken as the object identifier of the current tracking frame, otherwise the tracking result is determined to be empty, and the tracking target is not updated. When the current tracked object identifier is detected in a continuous preset frame number, the tracking of the object identifier is stopped. When the bounding box appears in the subsequent target video frame, a new object identifier is used to perform target tracking again.

[0059] In this way, by tracking the continuous frames of the target video frames, the driving trajectory of the same target can be obtained in real time, i.e., the position coordinate change information of the tracking target in the target video frames.

[0060] In the above embodiment, by tracking the bounding box of the target in the target video frames, the object identifier corresponding to each bounding box can be identified. When the bounding boxes correspond to the same object identifier, it can be considered that they are the same target. In this way, the tracking of the target can be accurately realized, and thus the driving trajectory of the target can be obtained.

[0061] In step S208, when it is determined that the current video frame in the target video frames meets the preset target condition based on the driving trajectory, marker recognition is performed based on the current video frame to obtain corresponding target marker information.

[0062] The preset target condition can be understood as a preset image clarity condition, and the preset image clarity condition can be constituted by at least one condition. Specifically, for the same target, the computer device can track the target in the target video frames in real time based on the form trajectory of the target. The computer device can judge the target video frames in the video stream in real time. When the current video frame meets the preset image clarity condition, the current video frame can be directly captured to obtain the target marker information based on the current video frame.

[0063] In one of the embodiments, the target can be a container vehicle, and the target marker information can be container marker information. For the same container vehicle currently tracked, the computer device can select a high-quality video frame from the target video frames to perform container marker recognition to obtain the container marker information of the container carried by the vehicle passing through the road portal.

[0064] In one of the embodiments, after the computer device selects a high-quality current video frame to perform marker recognition, the subsequent video frames with the same object identifier of the target can no longer be processed and are directly discarded until the object identifier is switched and step S208 is repeated, so as to reduce the computing pressure of the processor.

[0065] In one embodiment, the current video frame meeting the preset target condition can be a video frame with the clearest image within a preset number of previous frames. The current video frame meeting the preset target condition can also be a most stable video frame determined based on the previous video frames, etc. The current video frame meeting the preset target condition can also be a video frame with a bounding box area ratio greater than a preset proportion threshold, or a video frame with a bounding box edge distance to the edge of the video frame less than a preset distance, etc.

[0066] Further, the computer device can perform OCR (Optical Character Recognition) on the current video frame that meets the preset target condition to recognize the text information in the current video frame, and detect the position information of each text line. Thus, the container identification information appearing on the container in the current video frame is recognized. In one embodiment, the computer device can perform OCR recognition by using Differential Binarization and CRNN+CTC (Convolutional Recurrent Neural Network+Neural Network-based Time Sequence Classification) network.

[0067] In one embodiment, after the computer device performs OCR recognition on the current video frame to obtain the corresponding text information, the text information that meets the target mark feature, such as text length, text content, or text position, is taken as the target mark information.

[0068] The above target mark recognition method screens out the target video frame including the target by performing image detection processing on the video stream obtained by monitoring the target. Then, the target is accurately tracked based on the screened target video frame. Based on the tracking result, a clear image including the complete target can be accurately and reliably extracted from the video stream, text recognition is performed, and thus the target mark information is extracted. The present application can automatically capture a video frame meeting the preset target condition in the monitoring video stream scene without manual feature design, so as to accurately recognize the target mark based on the video frame. Moreover, the whole process is automatic detection and recognition, which greatly improves the target mark recognition efficiency.

[0069] In one embodiment, step S204, that is, performing image detection processing on the video frames in the video stream to screen out the video frames including the target as the target video frames, includes:

[0070] Step S302, performing target detection on each video frame in the video stream to obtain the candidate video frame including the target.

[0071] In some embodiments, the computer device can input each video frame in the video stream to the target detection model after pre-processing (such as adjusting the size to a specific size). The target detection model can be YOLO (You only look once, a target detection algorithm), YOLOv2, YOLOv3, SSD (Single Shot MultiBox Detector, single step multi-frame detection), Faster-RCNN (a neural network-based target detection algorithm), CenterNet (an object detection algorithm), etc. Then, whether the target is included in the video frame can be detected by the target detection model.

[0072] In some embodiments, the computer device can input each video frame into the target detection model after resizing, and output one or more detection boxes and attribute information of each detection box, such as coordinates, length and width, confidence, etc., by the target detection model. Then, for each video frame, the target detection model can select one of the multiple detection boxes as the bounding box according to a certain standard. In this embodiment, the certain standard can be the detection box with the largest area, which is used as the bounding box in the video frame to determine the target. In other embodiments, the detection box with the highest confidence can also be selected as the bounding box. The embodiments of the present application do not limit this.

[0073] For example, the computer device can use a target detection model composed of a YOLOv3 target detection network to input the video frame resized to 416x416. The confidence threshold is set to 0.5 (used to judge the confidence of the detected target belonging to a certain category, and if it is lower than the threshold, it is discarded), and the non-maximum suppression is set to 0.3 (used to process overlapping bounding boxes, if it is greater than the value, it is judged as the same target, and the redundant detection box is discarded), and the output result is one or more detection boxes containing the target detected by the target detection model, and the attribute information of the detection box, such as coordinates, length and width, confidence, etc. Then, the computer device can filter the above results, and select one of the multiple detection boxes as the main target according to a certain standard for each video frame, that is, to select the bounding box. In this embodiment, the certain standard can be the bounding box with the largest area, that is, the main target of the video frame.

[0074] It should be noted that in the road portal scene, when a container vehicle needs to be identified, since a portal generally only has the width of a lane allowing one vehicle to travel, in this case, the main target in the video frame can be defaulted as the target vehicle to be detected, even if the main target is not the target vehicle, it does not affect because the target integrity discrimination algorithm in the subsequent steps will also be discriminated and removed.

[0075] Step S304: input the candidate video frame into the target integrity discrimination model, process by the target integrity discrimination model, and output the target integrity discrimination result.

[0076] The target integrity discrimination model is used to determine whether the video frame of the detected target includes a complete target object. In an embodiment, when the target is a container vehicle and the target marker information is container marker information, the target integrity discrimination model can be a container integrity discrimination model. The container integrity discrimination model is used to determine whether the candidate video frame of the detected target vehicle includes a complete container, i.e., the container includes complete marker information.

[0077] In an embodiment, the computer device can use a classification network to build the target integrity discrimination model, such as a densenet (dense network), a vgg-net (computer vision network), a resnet (residual network), etc. The target integrity discrimination model is then trained by using training samples to obtain a trained target integrity discrimination model. The candidate video frame is classified by using the target integrity discrimination model to output a target integrity discrimination result.

[0078] In an embodiment, when the target is a container vehicle, the corresponding target integrity discrimination model can be a container integrity discrimination model. The computer device can use a classification network to build the container integrity discrimination model, such as a densenet (dense network), a vgg-net (computer vision network), a resnet (residual network), etc. The container integrity discrimination model is then trained by using training samples to obtain a trained container integrity discrimination model. The candidate video frame is classified by using the container integrity discrimination model to output a container integrity discrimination result. The container integrity discrimination result can include a result indicating a complete container and a result indicating an incomplete container.

[0079] Reference Figure 4 , Figure 4 The left side is a display diagram of a video frame including an incomplete container, Figure 4 The right side is a display diagram of a video frame including a complete container.

[0080] In an embodiment, the computer device can adjust the size of the bounding box in the candidate video frame to a preset size (for example, 224*224 size). After the adjustment, the image block corresponding to the bounding box is input into the target integrity discrimination model to perform target integrity discrimination. The output is complete or incomplete and target integrity confidence. In this way, the interference of other irrelevant information in the video frame can be removed, and the discrimination accuracy can be improved.

[0081] In some embodiments, the computer device can train the container integrity discrimination model in the following way: firstly, collect data (training samples and validation / test data): collect video data at the toll gate, use the target detection network to detect the main target (vehicle) and save the image block to the disk. Since the vehicle entering the toll gate is a continuous moving object in the video, when the vehicle enters the toll gate, the container is gradually complete from incomplete to complete and then to incomplete into the camera, that is, it is easy to determine the boundary point of the video frame of the incomplete (abnormal) and complete container (normal). The same reason applies to the vehicle exiting the toll gate. In this way, the video frame image can be quickly labeled as complete or incomplete (for this application, one side of the container with complete label information is considered complete). In addition, some other complete container images of vehicles can also be collected as positive samples and added to the training set. In addition to the incomplete container images collected, the positive sample images can also be randomly cropped to obtain negative samples (the container is cut off), so as to realize data enhancement and increase the sample space of the training set.

[0082] Further, the computer device can use resnet50 and train it through positive and negative samples, test the trained model using validation data, and end the training when the end condition is met. The end condition includes any one of the following conditions: 1) the accuracy of the trained container integrity discrimination model exceeds a certain threshold, for example, 98%; 2) the difference in the accuracy of the trained container integrity discrimination model is less than a threshold, for example, less than 0.1%, for two or more times in succession; 3) the difference in the loss function of the trained container integrity discrimination model is less than a threshold, for example, less than 0.1%, for two or more times in succession; 4) stop training after the number of times of training the model using all the training set reaches a certain threshold, and select the best model or the last trained model.

[0083] Step S306, screen the candidate video frame corresponding to the target integrity discrimination result indicating that the target includes a complete target as the target video frame.

[0084] Specifically, the computer device can screen the candidate video frame corresponding to the target integrity discrimination result indicating that the target includes a complete target as the target video frame.

[0085] In the above embodiments, the target detection is used to screen the candidate video frame including the target from the video stream. When the target is in a moving state, the detected target may not contain a complete marking pattern, so the target integrity of the detected target is identified, and the video frame including a complete target is screened as the video frame for subsequent tracking, which can ensure that the candidate is identified based on a complete marking pattern when identifying the marking, and improve the identification accuracy.

[0086] In one of the embodiments, when it is determined based on the driving track that the current video frame in the target video frame meets the preset target condition, the marker recognition is performed based on the current video frame to obtain the corresponding target marker information, including: determining, according to the driving track, the current video frame in the target video frame and a prequel frame before the current video frame; the prequel frame includes a first video frame before the current video frame and a second video frame before the first video frame; determining whether the current video frame meets the preset target condition according to target related information corresponding to the first video frame, the second video frame and the current video frame respectively; wherein the target related information includes at least one of position information corresponding to the target and target complete confidence; when the current video frame meets the preset target condition, performing marker recognition based on the current video frame to obtain the corresponding target marker information.

[0087] Specifically, the computer device can determine, according to the driving track of the target, the current video frame in the target video frame and a prequel frame before the current video frame. The prequel frame specifically includes a first video frame before the current video frame and a second video frame before the first video frame. The first video frame can be a previous frame or previous N frames of the current frame, and the second video frame can be a previous frame or previous N frames of the first video frame, which is not limited in the embodiments of the present application.

[0088] Further, the computer device can determine whether the current video frame meets the preset target condition according to target related information corresponding to the first video frame, the second video frame and the current video frame respectively; wherein the target related information includes at least one of position information corresponding to the target and target complete confidence. When the current video frame meets the preset target condition, marker recognition is performed based on the current video frame to obtain the corresponding target marker information.

[0089] In one of the embodiments, when the target is a container vehicle, the computer device can determine whether the current video frame meets the preset target condition according to container related information corresponding to the first video frame, the second video frame and the current video frame respectively; wherein the container related information includes at least one of position information corresponding to the container and container complete confidence; when the current video frame meets the preset target condition, container marker recognition is performed based on the current video frame to obtain the corresponding container marker information.

[0090] In one embodiment, the location information corresponding to the target includes preset location points for selecting the bounding box of the target vehicle. The preset target conditions include at least one of the following conditions: (condition 1) the angle between a first vector formed by the first preset location point corresponding to the first video frame and the second preset location point corresponding to the second video frame, and a second vector formed by the second preset location point and the third preset location point corresponding to the current video frame, is less than or equal to a preset angle threshold; (condition 2) the projection distance from the third preset location point to the second preset location point is greater than the distance from the first preset location point to the second preset location point; and (condition 3) the target integrity confidence corresponding to the current video frame is greater than or equal to a preset confidence threshold.

[0091] It should be noted that when a target is in motion, its trajectory can be obtained through tracking. However, if you want to capture images in real time to obtain stable and high-quality video frames, you need to determine whether to capture the image while the target is being tracked. In other words, you need to make a judgment on the current video frame and not wait until the trajectory is complete.

[0092] In the case of container recognition, the purpose of the scenario is to identify the markings on the container, and the cameras of the monitoring equipment located at the road checkpoint usually shoot from the back of the vehicle. When the vehicle appears completely in the camera area, the earlier it appears, the closer it is to the camera, so it needs to be captured as soon as possible and cannot be allowed to move further away.

[0093] To achieve more stable image capture, this application can determine whether to capture the current video frame by judging whether the current video frame meets at least one of the following three conditions. For example... Figure 5 The diagram illustrates the principle of the three-point capture method: ABCDE represents the center point positions of the target tracking across five frames, and A->B->C->D->E represents the trajectory of these five frames, meaning A appears earlier than B, B earlier than C, and so on. Specifically, the center point position can be the coordinates of the center point of the bounding box used to select the target in the video frame, as described in the previous embodiments, or other feature points of the bounding box, such as vertex endpoints. This embodiment does not limit this specific location.

[0094] Assuming the current video frame is at time C, the trajectory A->B->C already exists. Points D and E are future times and have not yet appeared. Therefore, point C is the third preset position point corresponding to the current video frame; point B is the first preset position point corresponding to the first video frame; and point A is the second preset position point corresponding to the second video frame. Taking the position information of points A, B, and C, we calculate the angle between the first vector AB and the second vector AC. This angle is less than θ. th Threshold (Condition 1); Draw a perpendicular line from point A to line AB. Figure 5the vertical distance (projection distance) from point C to Line A and the distance between points A and B, respectively, and denoted as d CA and d AB , d CA is greater than d AB (condition 2); Figure 5 the target complete confidence corresponding to point C is greater than a preset confidence threshold c th , or the target complete confidences corresponding to points A, B and C are all greater than a preset confidence threshold (condition 3). If at least one of conditions 1, 2 and 3 is satisfied, the current video frame is captured. It can be understood that when conditions 1, 2 and 3 are all satisfied, the image quality of the current video frame is better. The current video frame is the image frame in which point C is located. If not, the next time the frame corresponding to point D appears, the conditions 1, 2 and 3 are repeatedly calculated using the three-point trajectory B->C->D. The sequence with time or frame as the independent variable is used until the conditions are satisfied and then captured. After the current video frame satisfying the conditions is captured, the same tracking object identifier appearing later is no longer captured and discarded directly, so that each tracking object identifier is captured only once. In one embodiment, θ th is set to 15°, and c th is set to 0.9, but can also be set to other values, which are not limited in the present application.

[0095] In one embodiment, conditions 1 and 2 are used to ensure that the current tracking video frame is stable and clear, so as to facilitate subsequent container marking identification.

[0096] In one embodiment, for a scene photographed from the front by the camera, the three-point capture method (counting 3 frames from the next tracking object identifier appearing) can be applied to capture the video frame as late as possible to extract information, so as to extract the video frame satisfying the conditions as quickly and accurately as possible.

[0097] In the above embodiment, according to the target related information corresponding to the first video frame, the second video frame and the current video frame, it can be accurately and quickly judged whether the current video frame satisfies the preset target condition. Since the target related information includes at least one of the position information corresponding to the target and the target complete confidence, a high-quality current video frame can be selected from the video frames for target marking identification, and the marking identification accuracy can be greatly improved.

[0098] The following is taken as an example of a container vehicle and container number as container marking information. In one embodiment, the container marking information includes information of at least one information category, and the container marking identification is performed based on a current video frame to obtain corresponding container marking information, including: performing image text recognition on the current video frame to obtain a plurality of candidate text lines; for each candidate text line identified, determining the text length corresponding to the respective candidate text line; matching the text length corresponding to each candidate text line with a preset length, based on the information category corresponding to the matched preset length, adding the candidate text line to the candidate set of the corresponding information category; and from the candidate set corresponding to each information category respectively, screening out target text lines that meet the container marking characteristics, and obtaining corresponding container marking information based on the screened target text lines.

[0099] The information category can specifically include a check code, a box owner code, a registration code, and a box type code. The preset length corresponding to the check bit is a first value (such as a numerical value 1), the preset length corresponding to the box owner code and the box type code is a second value (such as a numerical value 4), the preset length corresponding to the registration code is a third value (such as a numerical value 6), and the preset length further includes a fourth value (such as a numerical value 7), which is the sum of the first value and the third value.

[0100] Specifically, the computer device can perform image text recognition on the current video frame to obtain a plurality of candidate text lines and the content corresponding to each candidate text line. For each text line identified, the text length of the text line is determined. When the text length of the candidate text line is the first value, the candidate text line is added to the candidate check code set corresponding to the first value. When the text length of the candidate text line is the second value, the box owner code table and the box type code table are respectively looked up and matched, when a match is successful in the box owner code table, the candidate text line is added to the candidate box owner code set, and when a match is successful in the box type code table, the candidate text line is added to the candidate box type code set. When the text length of the candidate text line is the third value and the text content is all numbers, the candidate text line is added to the candidate registration code set. When the text length of the candidate text line is the fourth value, the candidate text line is split, and based on the split result, the candidate set addition step corresponding to the first value and the candidate set addition step corresponding to the third value are respectively performed. When the text length of the candidate text line does not correspond to any of the above-mentioned preset lengths, the candidate text line is discarded. For each candidate text line, the candidate text line can be added to the corresponding candidate set in the above-mentioned manner.

[0101] Further, the computer device can screen out target text lines that meet the container marking characteristics from the candidate set corresponding to each information category respectively, and obtain corresponding container marking information based on the screened target text lines.

[0102] In one of the embodiments, from the candidate set corresponding to each information category respectively, the target text line meeting the container marking feature is screened out, and the corresponding container marking information is obtained based on the screened target text line, including: nested traversal is performed on the candidate owner code set corresponding to the owner code and the candidate registration code set corresponding to the registration code; the test check code is calculated based on the combination of each nested traversal, and the test check code is searched in the candidate check code set corresponding to the check code; when the test check code is found in the candidate check code set, and the candidate text line where the owner code and the container type code in the current traversal combination are located meets the collinear condition with the candidate text line corresponding to the test check code, the owner code in the current combination is taken as the target owner code, the registration code in the current combination is taken as the target registration code, and the test check code is taken as the target check code; the container type code in the candidate container type code set corresponding to the container type code, which meets the position corresponding relationship with the target owner code or the target registration code, is taken as the target container type code; the target owner code, the target registration code, the target check code and the target container type code are taken as the container marking information corresponding to the container.

[0103] It should be noted that the candidate text line where the owner code and the container type code in the current traversal combination are located meets the collinear condition with the candidate text line corresponding to the test check code, which can be specifically that the center points of the three codes in the candidate text line are collinear. The center points of the three codes in the candidate text line are collinear, and the area of the triangle formed by the first and last of the three center points can be directly calculated. If the area is less than a threshold A th , then the three points are collinear.

[0104] The container type code meets the position corresponding relationship with the target owner code or the target registration code, which can be specifically that the candidate text line where the container type code is located is below the text line where the target owner code is located, or the candidate text line where the container type code is located is below the text line where the target owner code is located.

[0105] In this way, from the candidate set corresponding to each information category respectively, the target text line meeting the container marking feature can be accurately and quickly screened out, and the corresponding container marking information can be obtained based on the screened target text line.

[0106] In one specific embodiment, referring to Figure 6 , Figure 6 is a flowchart of the container marking information extraction step based on the current video frame in one embodiment. As Figure 6 shown, the container marking information extraction step based on the current video frame can specifically include the following steps:

[0107] a) Select a candidate text line from the OCR recognition results. If its confidence score is greater than the threshold (e.g., 0.4), proceed to the next step; otherwise, discard it.

[0108] b) Remove spaces from the text recognition content.

[0109] c) Check the text length of the candidate text line. If it is equal to 1, 4, 6, or 7, proceed to the next step; otherwise, discard it.

[0110] d) If the text length is equal to 1 and the content is a number, then add it to the candidate checksum set.

[0111] e) If the text length is 4, check the box owner code table. If not found, continue searching in the box type code table. If not found either, search again. While searching, calculate the edit distance. If the replacement error in the edit distance is less than or equal to the preset value, such as less than or equal to 1, correct the error according to the preset error correction table. If the edit distance is zero after correction, the match is successful. If the box owner code is successfully matched, add it to the candidate box owner code set. If the box type code is matched, add it to the candidate box type code set. Otherwise, discard it.

[0112] f) If the text length is 6 and the content does not contain letters, add it to the candidate registration code set.

[0113] g) If the text length is 7, split it into 6-1 segments and then execute d) and f) separately.

[0114] h) Perform nested traversal on the obtained candidate box main code set and candidate registration code set, and calculate the test check code corresponding to the combination in the current traversal. If the calculated test check code is found in the candidate check code set, and the center points of the text lines containing the three codes are collinear, then the correct target box main code, target registration code and target check code have been found.

[0115] i) Check the position of the box type code in the current traversal, because the box type code may appear below the box owner code (a minority) or the registration code (most), such as Figure 7 As shown, Figure 7 This is a schematic diagram illustrating the possible positions of the container type code in a container marking in one embodiment. Therefore, we check either "distance from the top left corner to the bottom left corner of the registration code (AB) / distance from the top left corner of the registration code to the top left corner of the container type code (AC)" or "distance from the top left corner to the bottom left corner of the container owner code (DE) / distance from the top left corner of the container owner code to the top left corner of the container type code (DF)". If either of these values ​​is greater than a threshold, such as 0.7, then the target container type code is obtained.

[0116] In one embodiment, referring to Table 1 below, the algorithm for matching the container owner code and container type code described above can be explained in more detail:

[0117] Table 1. Logic diagram for matching container owner code and container type code.

[0118]

[0119] In the above embodiments, the text length corresponding to the candidate text line in the current video frame is matched with the preset length corresponding to each information category, so that the candidate text lines are added to the corresponding candidate sets respectively. This can quickly and accurately filter out the interference information with inconsistent word counts. Then, the target text lines that meet the container marking features are filtered out based on the candidate sets, which can improve both the filtering efficiency and the filtering accuracy.

[0120] In one specific embodiment, the target is a container vehicle, and the marking information is container marking information. (See reference...) Figure 8 , Figure 8 This is a flowchart illustrating a container marking recognition method in one embodiment. The proposed solution is a method for automatically recognizing container markings at checkpoints. The input is a video stream (frames) from a regular surveillance camera (not a snapshot camera), and the output is the identification of the container marking information for each passing vehicle, thereby automatically acquiring the container's marking information. Figure 8 As shown, taking a truck as the target vehicle as an example, the video stream is fed into the algorithm input interface frame by frame. First, a target detection algorithm is used to detect the truck. Then, the integrity of the container is judged on the detected truck image patch. Because it is a moving vehicle at the checkpoint, the detected truck may not necessarily contain a complete container. Therefore, the detection of whether the truck contains a complete container can be identified next. When the truck completely enters the camera's monitoring range, the truck is tracked as a single target. The best frame from the tracking results is selected as the container truck capture result, i.e., the truck image patch containing a complete container. Then, this image patch is sent to the OCR detection and recognition algorithm for text localization and content recognition. Finally, the verification and filtering algorithm extracts the container marking information. The embodiments of this application can track container trucks in the video stream and can automatically and reliably extract a clear image containing a complete container from the video stream for OCR recognition and mark extraction, improving efficiency. The OCR recognition of mark information in this application is applicable to more complex scenarios, such as tilt angle monitoring, and has error correction function.

[0121] It should be understood that, although Figure 2 , 3 The steps in flowcharts 6 and 8 are shown sequentially as indicated by the arrows; however, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order requirement for the execution of these steps, and they can be performed in other orders. Furthermore,Figure 2 , 3 At least some of the steps in 6 and 8 may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0122] In one embodiment, such as Figure 9 As shown, a target marker recognition device 900 is provided, including: an acquisition module 901, a filtering module 902, a target tracking module 903, and a marker recognition module 904, wherein:

[0123] The acquisition module 901 is used to acquire the video stream obtained from monitoring the target.

[0124] The filtering module 902 is used to perform image detection processing on video frames in the video stream and filter out video frames containing the target as target video frames.

[0125] The target tracking processing module 903 is used to perform target tracking processing based on the target video frame to obtain the target's driving trajectory.

[0126] The marker recognition module 904 is used to perform marker recognition based on the current video frame when the current video frame in the target video frame is determined to meet the preset target conditions based on the driving trajectory, and obtain the corresponding target marker information.

[0127] In one embodiment, the filtering module 902 is further configured to perform target detection on each video frame in the video stream to obtain candidate video frames containing targets; input the candidate video frames into the target integrity discrimination model, process them through the target integrity discrimination model, and output the target integrity discrimination result; and filter the candidate video frames corresponding to the target integrity discrimination result that represents the target containing the complete target as target video frames.

[0128] In one embodiment, the target tracking processing module 903 is further configured to obtain the position information of the bounding box used to select the target in the target video frame; perform target tracking processing based on the position information of the bounding box to obtain the object identifier corresponding to each bounding box; and determine the target's driving trajectory based on the position information of the bounding boxes corresponding to the same object identifier in each target video frame.

[0129] In one of the embodiments, the target identification module 904 is further configured to determine, according to the driving track, a current video frame and a previous frame of the target video frame; the previous frame includes a first video frame before the current video frame and a second video frame before the first video frame; determine whether the current video frame satisfies a preset target condition according to target related information corresponding to the first video frame, the second video frame and the current video frame respectively; the target related information includes at least one of position information corresponding to the target and target complete confidence; when the current video frame satisfies the preset target condition, perform target identification based on the current video frame to obtain corresponding target marking information.

[0130] In one of the embodiments, the position information corresponding to the container includes a preset position point for selecting a bounding box of the target vehicle, and the preset target condition includes at least one of the following conditions: an included angle between a first vector formed by a first preset position point corresponding to the first video frame and a second preset position point corresponding to the second video frame and a second vector formed by the second preset position point and a third preset position point corresponding to the current video frame is less than or equal to a preset angle threshold; a projection distance of the third preset position point to the second preset position point is greater than a distance from the first preset position point to the second preset position point; and the target complete confidence corresponding to the current video frame is greater than or equal to a preset confidence threshold.

[0131] In one of the embodiments, the target includes a container vehicle, and the marking information includes container marking information, the container marking information includes information of at least one information category, and the target identification module 904 is further configured to perform image text recognition on the current video frame to obtain a plurality of candidate text lines; for each candidate text line recognized, determine a text length corresponding to the corresponding candidate text line; match the text length corresponding to each candidate text line with a preset length, add the candidate text line to a candidate set of a corresponding information category based on an information category corresponding to the matching preset length, and select a target text line meeting a container marking feature from the candidate set corresponding to each information category respectively, and obtain corresponding container marking information based on the selected target text line.

[0132] In one of the embodiments, the information categories include the check code, the box owner code, the registration code and the box type code; the marking identification module 904 is further configured to perform nested traversal on the candidate box owner code set corresponding to the box owner code and the candidate registration code set corresponding to the registration code; calculate a test check code based on the combination of each nested traversal, and search for the test check code in the candidate check code set corresponding to the check code; when the test check code is found in the candidate check code set, and the candidate text line in which the box owner code and the box type code in the current traversal combination are located meets the collinearity condition with the candidate text line corresponding to the test check code, the box owner code in the current combination is taken as the target box owner code, the registration code in the current combination is taken as the target registration code, and the test check code is taken as the target check code; the box type code in the candidate box type code set corresponding to the box type code, which meets the position correspondence relationship with the target box owner code or the target registration code, is taken as the target box type code; and the target box owner code, the target registration code, the target check code and the target box type code are taken as the container marking information corresponding to the container.

[0133] The target marking identification device described above can realize accurate tracking of the target based on the target video frames screened out. Based on the tracking result, a clear image frame containing the complete target can be accurately and reliably extracted from the video stream, text recognition can be performed, and the target marking information can be extracted. The present application does not need manual feature design, and can automatically capture a video frame meeting the preset target condition in the monitoring video stream scene, so as to accurately identify the target marking based on the video frame. The whole process is automatic detection and identification, which greatly improves the target marking identification efficiency.

[0134] The specific limitations of the target marking identification device can be referred to the limitations of the target marking identification method described above, and will not be repeated here. Each module in the target marking identification device described above can be realized by software, hardware and combinations thereof in whole or in part. The modules described above can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0135] In one embodiment, a computer device, which can be a terminal or a server, is provided, and an internal structure diagram of the computer device can be as shown in Figure 10As shown in the figure. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium, an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is used to store container marking information. The network interface of the computer device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to implement a container marking identification method.

[0136] Those skilled in the art can understand that, Figure 10 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0137] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above method embodiments.

[0138] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps in the above method embodiments.

[0139] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned method embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0140] Any combination of the technical features in the above embodiments can be made, and for the sake of brevity, not all possible combinations are described above, however, as long as the combination of the technical features does not exist in contradiction, it shall be considered within the scope of the present disclosure.

[0141] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it shall not be understood as a limitation on the patent scope of the present application. It shall be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these shall be within the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A target marker identification method characterized by, The method comprises: acquiring a video stream obtained by monitoring a target; performing image detection processing on video frames in the video stream to screen video frames including the target from the video frames as target video frames; performing target tracking processing based on the target video frames to obtain a driving track of the target; determining a current video frame in the target video frames and a previous frame before the current video frame according to the driving track; the previous frame comprises a first video frame before the current video frame and a second video frame before the first video frame; determining whether the current video frame satisfies a preset target condition according to position information corresponding to the target corresponding to each of the first video frame, the second video frame and the current video frame; when the current video frame satisfies the preset target condition, performing marker recognition based on the current video frame to obtain corresponding target marker information; wherein the position information corresponding to the target comprises a preset position point of a bounding box for selecting a target vehicle, and the preset target condition comprises at least one of the following conditions: an included angle between a first vector formed by a first preset position point corresponding to the first video frame and a second preset position point corresponding to the second video frame and a second vector formed by the second preset position point and a third preset position point corresponding to the current video frame is less than or equal to a preset included angle threshold value; a projection distance of the third preset position point to the second preset position point is greater than a distance from the first preset position point to the second preset position point.

2. The method of claim 1, wherein, The image detection processing on the video frames in the video stream to screen the video frames including the target from the video frames as the target video frames comprises: performing target detection on each video frame in the video stream to obtain candidate video frames including the target; inputting the candidate video frames into a target integrity discrimination model to perform processing through the target integrity discrimination model to output target integrity discrimination results; screening the candidate video frames corresponding to the target integrity discrimination results indicating that the target is complete as the target video frames.

3. The method of claim 1, wherein, The target tracking processing based on the target video frames to obtain the driving track of the target comprises: acquiring position information of a bounding box for selecting the target in the target video frames; performing target tracking processing based on the position information of the bounding box to obtain object identifiers corresponding to each of the bounding boxes; determining the driving track of the target based on position information of the bounding boxes corresponding to the same object identifier in each of the target video frames.

4. The method of claim 1, wherein, The target comprises a container vehicle, the marker information comprises container marker information, the container marker information comprises information of at least one information category, and the marker recognition based on the current video frame to obtain the target marker information comprises: performing image text recognition on the current video frame to obtain a plurality of candidate text lines; for each of the recognized candidate text lines, determining a text length corresponding to the corresponding candidate text line; The text length corresponding to each candidate text line is matched with the preset length, and based on the information category corresponding to the matched preset length, the candidate text line is added to the candidate set of the corresponding information category; From the candidate set corresponding to each information category respectively, a target text line meeting the container marking feature is screened out, and corresponding container marking information is obtained based on the screened target text line.

5. The method of claim 4, wherein, The information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a box owner code, a registration code, and a box type code; the information category includes a check code, a ​ ​ ​ ​ ​ 6. A target marker identification apparatus characterized by comprising: ​ ​ ​ ​ The marking recognition module is configured to determine a current video frame in the target video frames and a previous frame before the current video frame according to the driving track; the previous frame includes a first video frame before the current video frame and a second video frame before the first video frame; determine whether the current video frame satisfies a preset target condition according to position information corresponding to the target corresponding to the first video frame, the second video frame and the current video frame respectively; when the current video frame satisfies the preset target condition, perform marking recognition based on the current video frame to obtain corresponding target marking information; wherein the position information corresponding to the target includes a preset position point for selecting a bounding box of a target vehicle, and the preset target condition includes at least one of the following conditions: an included angle between a first vector formed by a first preset position point corresponding to the first video frame and a second preset position point corresponding to the second video frame, and a second vector formed by the second preset position point and a third preset position point corresponding to the current video frame, is less than or equal to a preset angle threshold; a projection distance of the third preset position point to the second preset position point is greater than a distance from the first preset position point to the second preset position point. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 5.

Images