Accident vehicle association identification method and system based on video stream segmentation

Abstract

The invention discloses an accident vehicle association identification method and system based on video stream segmentation, and relates to the technical field of traffic management, and the method comprises the steps: generating a video retrieval constraint based on a first accident scene; establishing a traffic video information base; the traffic video information base is interacted, calling and screening are carried out according to the video retrieval constraint, and screened video information is obtained; based on a video stream segmentation algorithm, segmenting the screened video information to obtain segmented stream data; analyzing and processing the segmented stream data to generate a vehicle position information set; and performing relevance evaluation on the multiple groups of vehicle position information in the vehicle position information set to obtain a relevance evaluation result, and performing relevance identification on the accident vehicle according to the relevance evaluation result. Therefore, the technical effects of improving the multi-target tracking performance and performing correlation analysis on non-contact accidents are achieved.

Description

Technical Field

[0001] The invention relates to the technical field of traffic management, and in particular to an accident vehicle association identification method and system based on video stream segmentation. Technical Background

[0002] With the development of economy and society, the demand for efficient and accurate management of traffic accidents is increasing. Existing road and traffic monitoring systems widely use video cameras to monitor traffic flow and record accidents, and provide assistance for accident vehicle identification and marking. Existing technologies have difficulties in multi-target tracking and lack the ability to perform correlation analysis on non-contact accidents. Summary of the invention

[0003] The purpose of this application is to provide a method and system for associating and identifying accident vehicles based on video stream segmentation, so as to solve the technical problems of the difficulty of multi-target tracking and the lack of the ability to perform associative analysis on non-contact accidents in the prior art.

[0004] In view of the above technical problems, the present application provides an accident vehicle association identification method and system based on video stream segmentation.

[0005] In a first aspect, the present application provides an accident vehicle association identification method based on video stream segmentation, wherein the method comprises:

[0006] Acquire a first accident scene, and based on the first accident scene, generate a video retrieval constraint, wherein the video retrieval constraint includes a video time constraint and a video position constraint; establish a traffic video information library, wherein the traffic video in the traffic video information library has a position tag and a time tag; interact with the traffic video information library, call and filter according to the video retrieval constraint, and obtain filtered video information, wherein the filtered video information includes N video stream information, and N is greater than or equal to 1; based on a video stream segmentation algorithm, segment the filtered video information to obtain segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences; analyze and process the segmented stream data to generate a vehicle position information set, wherein the vehicle position information set includes K groups of vehicle position information; perform correlation evaluation on multiple groups of vehicle position information in the vehicle position information set, obtain correlation evaluation results, and associate the accident vehicle according to the correlation evaluation results.

[0007] In a second aspect, the present application further provides an accident vehicle association identification system based on video stream segmentation, wherein the system comprises:

[0008] A retrieval constraint module, the retrieval constraint module is used to obtain a first accident scene, and based on the first accident scene, generate a video retrieval constraint, wherein the video retrieval constraint includes a video time constraint and a video position constraint; a video information acquisition module, the video information acquisition module is used to establish a traffic video information library, and the traffic video in the traffic video information library has a position tag and a time tag; a retrieval call module, the retrieval call module is used to interact with the traffic video information library, call and filter according to the video retrieval constraint, and obtain filtered video information, wherein the filtered video information includes N video stream information, and N is greater than or equal to 1; a video stream segmentation module, the video stream segmentation module is used to segment the filtered video information based on a video stream segmentation algorithm, and obtain segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences; a position analysis module, the position analysis module is used to analyze and process the segmented stream data, and generate a vehicle position information set, and the vehicle position information set includes K groups of vehicle position information; an association evaluation module, the association evaluation module is used to evaluate the association of multiple groups of vehicle position information in the vehicle position information set, obtain the association evaluation result, and associate the accident vehicle according to the association evaluation result.

[0009] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0010] By acquiring the first accident scene, based on the first accident scene, a video retrieval constraint is generated, wherein the video retrieval constraint includes a video time constraint and a video position constraint; a traffic video information library is established, wherein the traffic video in the traffic video information library has a position tag and a time tag; the interactive traffic video information library is called and filtered according to the video retrieval constraint to obtain the filtered video information, wherein the filtered video information includes N video stream information, and N is greater than or equal to 1; based on the video stream segmentation algorithm, the filtered video information is segmented to obtain the segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences; the segmented stream data is analyzed and processed to generate a vehicle position information set, wherein the vehicle position information set includes K groups of vehicle position information; the correlation evaluation is performed on multiple groups of vehicle position information in the vehicle position information set, the correlation evaluation results are obtained, and the association identification of the accident vehicle is performed according to the correlation evaluation results. Thus, the technical effect of improving the multi-target tracking performance and realizing the correlation analysis of non-contact accidents is achieved.

[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly illustrate the technical means of the present application and to implement it in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The embodiments of the present invention and the following brief description are illustrated in conjunction with the accompanying drawings, which are described as follows:

[0013] Figure 1 This is a flow chart of the accident vehicle association identification method based on video stream segmentation of this application;

[0014] Figure 2 A schematic diagram of the process of obtaining and screening video information in the accident vehicle association identification method based on video stream segmentation in this application;

[0015] Figure 3 This is a structural diagram of the accident vehicle association identification system based on video stream segmentation in this application.

[0016] Explanation of the accompanying drawings: retrieval constraint module 11, video information acquisition module 12, retrieval call module 13, video stream segmentation module 14, location analysis module 15, association evaluation module 16. DETAILED DESCRIPTION

[0017] The present application solves the technical problems faced by the prior art of difficulty in multi-target tracking and lack of ability to perform association analysis on non-contact accidents by providing an accident vehicle association identification method and system based on video stream segmentation.

[0018] The solution in the embodiment of the present technology, in order to solve the above problems, adopts the following overall ideas:

[0019] First, the first accident scene is interacted, and the video time constraint and the video position constraint are obtained and set as video retrieval constraints; then, a traffic video information library is established, and the traffic video in the traffic video information library has a position tag and a time tag; the traffic video information library is interacted, and the filtering is called according to the video retrieval constraint to obtain the filtered video information; based on the video stream segmentation algorithm, the filtered video information is segmented to obtain the segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences; the segmented stream data is analyzed and processed to generate a vehicle position information set, and the vehicle position information set includes K groups of vehicle position information; the correlation evaluation is performed on multiple groups of vehicle position information in the vehicle position information set, and the correlation evaluation results are obtained, and the accident vehicle is associated with the identification according to the correlation evaluation results. Then, the technical effect of improving the multi-target tracking performance and realizing the correlation analysis of non-contact accidents is achieved.

[0020] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the drawings and specific implementation methods of the specification. It should be noted that the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should also be noted that for the sake of ease of description, only the parts related to the present invention are shown in the drawings, rather than all of them.

[0021] Embodiment 1

[0022] like Figure 1 As shown, the present application provides an accident vehicle association identification method based on video stream segmentation, the method comprising:

[0023] S100: Acquire a first accident scene, and generate a video retrieval constraint based on the first accident scene, wherein the video retrieval constraint includes a video time constraint and a video position constraint;

[0024] Optionally, the first accident scene includes occurrence time information and occurrence location confidence information of the target accident, and the determination of the video retrieval constraints can be achieved based on the occurrence time information and the occurrence location confidence information.

[0025] Optionally, the video time constraint is used to determine the time period within which the video is retrieved. Exemplarily, it includes a start time and an end time to narrow the search scope to the time period in which the accident occurred, which is determined based on the time of occurrence of the first accident scene. The video location constraint is used to determine the location or area within which the video is retrieved. By determining a geographic location or spatial coordinates, it is ensured that the retrieved video matches the location of the first accident scene.

[0026] S200: Establishing a traffic video information database, wherein the traffic videos in the traffic video information database have position tags and time tags;

[0027] The traffic video information database is a database containing traffic surveillance video information of the area or region where the accident occurred, and is used to provide a retrieval space for the retrieval of the target accident vehicle video information.

[0028] Optionally, the traffic video information database is a cloud database set up on a remote server. It is built based on big data technology. The remote server is a remotely connected cloud storage device, which can be an independent storage device, or a distributed storage device or other type of computing device interconnected through a network. Using a remote server to establish a traffic video information database reduces deployment costs, has good maintainability, stability, and security, and is easy to expand and configure.

[0029] Exemplarily, a traffic video information library is established. First, data collection is performed to collect traffic video data, wherein the sources of the video data include traffic cameras, surveillance cameras, driving recorders, network videos, etc. And the video has an accurate location tag (geographic coordinates) and a time tag. This information is associated with each video file. Next, the above video data is processed to ensure quality and consistency. Including correcting timestamps, handling video quality issues, etc. Then, a database design is performed to create a database for storing relevant information of traffic videos. The database contains data tables to store metadata of the video, such as video ID, location information, time information, file path, etc. Then, the data is imported into the database, and an index is established to improve the database's processing capabilities for rapid query and retrieval. Finally, a data interface for accessing and retrieving is set for the database, while ensuring the security and privacy of the retrieval.

[0030] Optionally, based on real-time monitoring data, the traffic video information database is updated in real time to ensure that it contains the latest traffic video data.

[0031] Optionally, based on the privacy and security regulations of the target area, set the retrieval permissions and retrieval rules of the traffic video information database to ensure compliance with data collection and processing and protect video data involving personal information.

[0032] S300: interacting with the traffic video information database, calling and filtering according to the video retrieval constraint, and obtaining filtered video information, wherein the filtered video information includes N video stream information, and N is greater than or equal to 1;

[0033] Among them, the N video stream information in the screened video information corresponds to videos from N sources that are compatible with the video time constraint and the video position constraint. The N video stream information reflects the target accident vehicle and target accident monitoring information from different monitoring devices and different angles, which improves the breadth and difference of the acquired video information, and provides data support for video stream segmentation and association analysis identification from multiple angles, thereby improving the accuracy and robustness of the accident vehicle association identification.

[0034] Further, such as Figure 2 As shown, the traffic video information library is interacted, and the video retrieval constraint is used to perform a call and filter, and the filtered video information is obtained. Step S300 includes:

[0035] Based on the video retrieval constraint, obtaining the video time constraint and the video position constraint;

[0036] Based on the first accident scenario, obtaining a video clarity constraint;

[0037] According to the video time constraint, the video position constraint, and the video clarity constraint, traverse and filter to obtain the filtered video information.

[0038] The video clarity constraint refers to the maximum clarity threshold that satisfies the subsequent segmentation process, including resolution, frame rate or other video quality constraints. The video clarity constraint is used to ensure that the screened video information is clear and usable.

[0039] Optionally, the video clarity constraint is a conditional constraint, and by dividing the environmental conditions of the first accident scene, different levels and clarity constraint thresholds are determined to improve the adaptability of the video clarity constraint to different environments and retain the available video information to the greatest extent. Exemplarily, for the first accident scene with good ambient light and ambient visibility, the video clarity constraint is stricter to ensure that good quality video information is obtained. If the ambient light, ambient visibility and other environmental conditions of the first accident scene are not good, the clarity constraint threshold corresponding to the lower level of the environmental conditions is selected to ensure the number of samples of the obtained video information.

[0040] Optionally, based on an image quality analysis operator or an image quality analysis tool, an analysis and evaluation of the video clarity is performed, such as SSIM (Structural Similarity Index), etc., to achieve comparison with the video clarity constraint.

[0041] Optionally, if it is not possible to obtain or a sufficient amount of video information under a certain video clarity constraint, the video clarity constraint is updated to lower the clarity requirement of the video clarity constraint to ensure the acquisition of the filtered video information.

[0042] S400: Segmenting the screened video information based on a video stream segmentation algorithm to obtain segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences;

[0043] Optionally, the K groups of foreground image sequences in the segmented stream data refer to vehicle images appearing in the first accident scene, and each group of foreground image sequences corresponds to an individual vehicle.

[0044] Optionally, the lengths of the K groups of foreground image sequences in the segmented stream data are different, that is, the lengths of time and the order of appearance of multiple vehicles in the accident scene are different. In addition, the K groups of foreground image sequences have the same time coordinate scale and position coordinate reference, which facilitates the registration of the time scale and spatial scale of the foreground image sequences of different groups, so that the K groups of foreground images can be analyzed and compared.

[0045] Further, based on the video stream segmentation algorithm, the screened video information is segmented to obtain segmented stream data, and step S400 includes:

[0046] Based on the filtered video information, obtaining first video stream information;

[0047] Setting frame extraction control parameters, and performing frame extraction processing on the first video stream information based on the frame extraction control parameters to obtain a first image sequence, wherein the first image sequence includes a plurality of frame images;

[0048] Acquire a first frame image in the first image sequence, perform multi-target vehicle detection on the first frame image through a vehicle detection channel, acquire a multi-target vehicle detection result, and store it in a multi-target vehicle detection result set;

[0049] Acquire the second frame image ... the Mth frame image in the first image sequence, perform multi-target vehicle detection on the second frame image ... the Mth frame image through a vehicle detection channel, acquire the multi-target vehicle detection result, and store it in the multi-target vehicle detection result set;

[0050] According to the multi-target vehicle detection result set, the first image sequence is segmented into video streams to obtain the segmented stream data.

[0051] Optionally, the first video stream information is video stream information from any angle or any source in the filtered video information, reflecting the first accident scene information from a single perspective. By traversing and filtering multiple video stream information in the video information, multi-angle and all-round segmentation processing of the accident scene information is achieved.

[0052] Optionally, based on the filtered video information, the video stream is segmented, and the video stream segmentation result includes a plurality of segmented stream data. The plurality of segmented stream data are used for subsequent position acquisition result verification, thereby improving the accuracy and robustness of the accident vehicle association identification.

[0053] Optionally, frame extraction refers to the process of extracting frames from video stream information, which is used to compress and streamline information, facilitate data transmission and processing, and reduce transmission load and computing power consumption. Frame extraction includes a key frame-based frame extraction method and a key frame extraction method based on frame extraction parameters.

[0054] Exemplarily, the compressed video stream information includes I frame (Intra Frame), P frame (Predicted Frame), and B frame (Bidirectional Frame). Among them, I frame is a key frame in video encoding. Each I frame is a complete and independent image frame, which is not dependent on the information of other frames for encoding, and maintains the integrity of image quality. P frame is a predicted frame by motion estimation and compensation for the forward reference frame. P frame only stores the difference or residual information between the reference frame and the reference frame, thereby achieving image compression. P frame depends on the previous I frame or P frame for decoding, and performs motion compensation with the reference frame to restore the image. B frame is a bidirectional predicted frame by motion estimation and compensation for the forward and backward reference frames. B frame stores the difference or residual information between the previous and next frames, and can achieve a higher degree of compression. B frame needs to refer to the previous and next frames for motion compensation when decoding. The compressed video stream information includes its own key frame interval. The key frame interval is used as the frame extraction parameter to extract the I frame and obtain an image frame containing multiple vehicle initial positions.

[0055] Exemplarily, for the video stream information of the parameters that have not been compressed or position compressed, the frame extraction control parameter is set based on the frame rate of the video stream information, wherein the frame extraction control parameter includes the frame extraction frame number interval or the frame extraction time interval.

[0056] Optionally, the vehicle detection channel is used to detect and identify multiple vehicles in an image sequence. The vehicle detection channel is built based on computer vision technology, and the models used include YOLO (You Only Look Once) or Faster R-CNN. Through various target detection models based on convolutional neural networks, end-to-end detection of multiple vehicles is achieved, the location range of multiple vehicles is reduced, and the subsequent video stream segmentation is facilitated, the computing power consumption of the segmentation operation is reduced, and the efficiency of video stream segmentation is improved. In addition, the acquisition of multi-target vehicle detection result sets facilitates the parallel processing of video stream segmentation, further improving the processing efficiency.

[0057] Further, according to the multi-target vehicle detection result, the first image sequence is segmented into video streams to obtain segmented stream data, and the step further includes:

[0058] Activate the image segmentation channel, take the multi-target vehicle detection result set as input, and obtain a multi-target vehicle segmentation image set;

[0059] Performing feature extraction on the multi-target vehicle segmentation image set to obtain a multi-target vehicle feature set, wherein the multi-target vehicle feature set includes vehicle features of K vehicles;

[0060] A cluster analyzer is constructed to perform cluster analysis based on the multi-target vehicle feature set to obtain the segmented stream data, wherein the segmented stream data includes K groups of segmented stream data sets corresponding to the features of K vehicles in the video stream information.

[0061] Optionally, the image segmentation channel is used to obtain a foreground image set containing multiple vehicle images according to the multi-target vehicle detection result set. The image segmentation channel involves an image segmentation algorithm, including image segmentation based on edge monitoring, image segmentation based on Mask R-CNN, and the like.

[0062] Optionally, the image segmentation channel performs pre-segmentation based on multiple anchor_boxes in the multi-target vehicle detection result set, and divides multiple frame images in the multi-target detection result set into multiple ROI images. Among them, the ROI image is a partial image of the frame image, and the multiple ROI images contain the vehicle image information required for segmentation. Through pre-segmentation, the frame image with high resolution and large amount of information is converted into an ROI image with clear targets and small amount of information, which is convenient for subsequent processing and improves the processing efficiency and image segmentation channel processing efficiency.

[0063] Optionally, the image segmentation image further includes a preprocessing end for preprocessing the multi-target vehicle detection result set to improve image quality and reduce noise. The preprocessing steps include image denoising, image enhancement, color correction, etc. to improve the usability of the image.

[0064] Optionally, feature extraction of the multi-target vehicle segmentation image set involves identification information such as color, size, shape, texture, and license plate of the target vehicle. Feature extraction methods used for multiple features include color histogram, shape information, gray level co-occurrence matrix (GLCM) statistical features, local binary pattern (LBP) features, scale-invariant feature transform (SIFT), deep learning feature extraction methods based on convolutional neural networks, etc.

[0065] Optionally, the multi-target vehicle feature set is used for target recognition, tracking, and classification. Based on the multi-target vehicle feature set, feature clustering analysis is performed to divide the multi-target vehicle segmentation image set into K groups of segmentation stream data sets corresponding to K vehicles in the video stream information. The problem of low recognition accuracy caused by vehicle position change and mutual occlusion in image segmentation and target monitoring is solved, the confidence and robustness of video stream segmentation are improved, and it has good anti-interference ability.

[0066] S500: Analyze and process the segmented stream data to generate a vehicle position information set, wherein the vehicle position information set includes K groups of vehicle position information;

[0067] Further, the segmented stream data is analyzed and processed to generate a vehicle position information set, wherein the vehicle position information set includes K groups of vehicle position information. Step S500 includes:

[0068] Build a vehicle tracker based on object tracking algorithm;

[0069] Randomly selecting from the segmented stream data to extract a first foreground image sequence;

[0070] Inputting the first foreground image sequence into the vehicle tracker to obtain first vehicle position information;

[0071] The remaining foreground image sequences in the segmented stream data are traversed and extracted, the first foreground image sequence is updated, and the vehicle position information set is acquired based on the updated first foreground image sequence.

[0072] Optionally, the vehicle tracker is constructed based on an object tracking algorithm, including a Kalman filter, a multi-object tracker (MOT), a particle filter, a mean shift, etc. Exemplarily, first, the vehicle tracking algorithm is initialized, a unique identifier is assigned to each vehicle, and its current position is recorded, then the vehicle tracker tracks the vehicle in continuous image frames, obtains a continuous sequence of vehicle position coordinates, and obtains vehicle position information.

[0073] S600: performing correlation evaluation on multiple groups of vehicle position information in the vehicle position information set, obtaining correlation evaluation results, and performing correlation identification of the accident vehicle according to the correlation evaluation results.

[0074] Further, a correlation evaluation is performed on a plurality of groups of vehicle position information in the vehicle position information set, a correlation evaluation result is obtained, and a correlation identification of the accident vehicle is performed according to the correlation evaluation result. Step S600 includes:

[0075] Obtain the location information of the accident vehicle;

[0076] Traversing the multiple sets of vehicle position information, pairing them with the accident vehicle position information in pairs, and obtaining multiple sets of paired vehicle position information;

[0077] Performing relative speed analysis on the plurality of groups of paired vehicle position information moment by moment to obtain a relative speed sequence;

[0078] Performing relative distance analysis on the plurality of groups of paired vehicle position information moment by moment to obtain a relative distance sequence;

[0079] Based on the correlation evaluation model, correlation evaluation is performed according to the relative speed sequence and the relative distance sequence to obtain the correlation evaluation result.

[0080] The accident vehicle refers to the vehicle that is known to have had an accident or a vehicle that has reported an accident in the first accident scene. By pairing multiple sets of vehicle location information with the accident vehicle location information, multiple sets of paired vehicle location information are obtained, ensuring the completeness of vehicle association analysis.

[0081] Optionally, the correlation evaluation model is used to perform correlation analysis on multiple sets of paired vehicle position information according to the relative speed sequence and the relative distance sequence. The construction of the correlation evaluation model is based on statistical analysis, correlation analysis or machine learning methods.

[0082] Exemplarily, a machine learning method is used in combination with supervised learning to construct a correlation evaluation model for non-contact accidents between vehicles. First, data about non-contact accidents between vehicles. These data include vehicle location information, speed information, timestamps, etc. Then, each set of data is marked as "related" or "irrelevant" as a supervised training label for the model. Next, the data is preprocessed, including data cleaning, feature extraction, and feature engineering. To calculate and obtain features such as relative speed, relative distance, acceleration, etc. Then, the data is divided into a training set and a test set, an appropriate machine learning algorithm is selected, and the training set is used to train the selected machine learning model. Optional models include logistic regression, support vector machine, decision tree, random forest, etc. Finally, the test set is used to evaluate the performance of the model. The performance parameters of the model include indicators such as precision, recall, F1 score, etc. According to the evaluation results, the hyperparameters of the model are adjusted to obtain better performance.

[0083] In summary, the accident vehicle association identification method based on video stream segmentation provided by the present invention has the following technical effects:

[0084] By acquiring the first accident scene, based on the first accident scene, a video retrieval constraint is generated, wherein the video retrieval constraint includes a video time constraint and a video position constraint; a traffic video information library is established, wherein the traffic video in the traffic video information library has a position tag and a time tag; the interactive traffic video information library is called and filtered according to the video retrieval constraint to obtain the filtered video information, wherein the filtered video information includes N video stream information, and N is greater than or equal to 1; based on the video stream segmentation algorithm, the filtered video information is segmented to obtain the segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences; the segmented stream data is analyzed and processed to generate a vehicle position information set, wherein the vehicle position information set includes K groups of vehicle position information; the correlation evaluation is performed on multiple groups of vehicle position information in the vehicle position information set, the correlation evaluation results are obtained, and the association identification of the accident vehicle is performed according to the correlation evaluation results. Thus, the technical effect of improving the multi-target tracking performance and realizing the correlation analysis of non-contact accidents is achieved.

[0085] Embodiment 2

[0086] Based on the same concept as the accident vehicle association identification method based on video stream segmentation in the above embodiment, Figure 3 As shown, the present application also provides an accident vehicle association identification system based on video stream segmentation, the system comprising:

[0087] A retrieval constraint module 11 is used to obtain a first accident scene, and generate a video retrieval constraint based on the first accident scene, wherein the video retrieval constraint includes a video time constraint and a video position constraint;

[0088] The video information acquisition module 12 is used to establish a traffic video information database, in which the traffic videos have position marks and time marks;

[0089] A retrieval calling module 13 is used to interact with the traffic video information library, perform calling and filtering according to the video retrieval constraint, and obtain filtered video information, wherein the filtered video information includes N video stream information, and N is greater than or equal to 1;

[0090] A video stream segmentation module 14 is used to segment the screened video information based on a video stream segmentation algorithm to obtain segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences;

[0091] A position analysis module 15, configured to analyze and process the segmented stream data to generate a vehicle position information set, wherein the vehicle position information set includes K groups of vehicle position information;

[0092] The association evaluation module 16 is used to evaluate the association of multiple groups of vehicle position information in the vehicle position information set, obtain the association evaluation result, and perform association identification of the accident vehicle according to the association evaluation result.

[0093] Furthermore, the search and call module 13 also includes:

[0094] An event constraint unit, configured to obtain the video time constraint and the video position constraint based on the video retrieval constraint;

[0095] A quality constraint unit, configured to obtain a video clarity constraint based on the first accident scene;

[0096] The screening and matching unit is used to traverse and screen according to the video time constraint, the video position constraint, and the video clarity constraint to obtain the screened video information.

[0097] Furthermore, the video stream segmentation module 14 also includes:

[0098] A segmentation object acquisition unit, configured to acquire first video stream information based on the screened video information;

[0099] A frame extraction unit, used to set frame extraction control parameters, and perform frame extraction processing on the first video stream information based on the frame extraction control parameters to obtain a first image sequence, wherein the first image sequence includes a plurality of frame images;

[0100] A first-frame vehicle detection unit is used to obtain a first frame image in the first image sequence, perform multi-target vehicle detection on the first frame image through a vehicle detection channel, obtain a multi-target vehicle detection result, and store it in a multi-target vehicle detection result set;

[0101] a continuous vehicle detection unit, configured to obtain a second frame image ... an M-th frame image in the first image sequence, perform multi-target vehicle detection on the second frame image ... an M-th frame image through a vehicle detection channel, obtain the multi-target vehicle detection result, and store it in the multi-target vehicle detection result set;

[0102] The video segmentation unit is used to perform video stream segmentation on the first image sequence according to the multi-target vehicle detection result set to obtain the segmented stream data.

[0103] Furthermore, the video segmentation unit also includes:

[0104] A segmented image acquisition unit is used to activate the image segmentation channel, take the multi-target vehicle detection result set as input, and acquire a multi-target vehicle segmentation image set;

[0105] A feature extraction unit is used to extract features from the multi-target vehicle segmentation image set to obtain a multi-target vehicle feature set, wherein the multi-target vehicle feature set includes vehicle features of K vehicles;

[0106] A clustering unit is used to construct a cluster analyzer, perform cluster analysis based on the multi-target vehicle feature set, and obtain the segmented stream data, wherein the segmented stream data includes K groups of segmented stream data sets corresponding to the features of K vehicles in the video stream information.

[0107] Furthermore, the location analysis module 15 also includes:

[0108] A tracker building unit, used for building a vehicle tracker based on an object tracking algorithm;

[0109] A tracking target acquisition unit, used for randomly selecting from the segmented stream data to extract a first foreground image sequence;

[0110] A first position acquisition unit, used for inputting the first foreground image sequence into the vehicle tracker to acquire first vehicle position information;

[0111] The position set acquisition unit is used to traverse and extract the remaining foreground image sequences in the segmented stream data, update the first foreground image sequence, and acquire the vehicle position information set based on the updated first foreground image sequence.

[0112] Furthermore, the association evaluation module 16 also includes:

[0113] A target vehicle information acquisition unit, used to acquire the location information of the accident vehicle;

[0114] A pairing unit, used for traversing the multiple sets of vehicle position information, pairing them with the accident vehicle position information in pairs, and obtaining multiple sets of paired vehicle position information;

[0115] A relative speed analysis unit, used to perform relative speed analysis on the plurality of sets of paired vehicle position information moment by moment to obtain a relative speed sequence;

[0116] A relative distance analysis unit, used to perform relative distance analysis on the plurality of sets of paired vehicle position information moment by moment to obtain a relative distance sequence;

[0117] The association evaluation unit is used to perform association evaluation based on the association evaluation model according to the relative speed sequence and the relative distance sequence to obtain the association evaluation result.

[0118] It should be understood that the embodiments mentioned in this specification focus on their differences from other embodiments. The specific embodiments in the aforementioned embodiment one are also applicable to the accident vehicle association identification system based on video stream segmentation described in embodiment two. For the sake of brevity of the specification, they will not be further elaborated here.

[0119] It should be understood that the embodiments disclosed in this application and the above description can enable those skilled in the art to use this application to implement this application. At the same time, this application is not limited to the above-mentioned embodiments, and obvious modifications and variations of the embodiments mentioned in this application also fall within the scope of the principles of this application.

Claims

1. An accident vehicle association identification method based on video stream segmentation, characterized in that: The method comprises: Acquire a first accident scene, and generate a video retrieval constraint based on the first accident scene, wherein the video retrieval constraint includes a video time constraint and a video position constraint; Establishing a traffic video information database, wherein the traffic videos in the traffic video information database have position tags and time tags; Interacting with the traffic video information database, calling and filtering according to the video retrieval constraint, and obtaining filtered video information, wherein the filtered video information includes N video stream information, and N is greater than or equal to 1; Based on a video stream segmentation algorithm, the screened video information is segmented to obtain segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences; Analyze and process the segmented stream data to generate a vehicle position information set, wherein the vehicle position information set includes K groups of vehicle position information; A correlation evaluation is performed on multiple groups of vehicle position information in the vehicle position information set, a correlation evaluation result is obtained, and a correlation mark of the accident vehicle is performed according to the correlation evaluation result.

2. The method according to claim 1, characterized in that Interacting with the traffic video information database, calling and filtering according to the video retrieval constraints, and obtaining filtered video information, including: Based on the video retrieval constraint, obtaining the video time constraint and the video position constraint; Based on the first accident scenario, obtaining a video clarity constraint; According to the video time constraint, the video position constraint, and the video clarity constraint, traverse and filter to obtain the filtered video information.

3. The method according to claim 2, characterized in that Based on the video stream segmentation algorithm, the screened video information is segmented to obtain segmented stream data, including: Based on the filtered video information, obtaining first video stream information; Setting frame extraction control parameters, and performing frame extraction processing on the first video stream information based on the frame extraction control parameters to obtain a first image sequence, wherein the first image sequence includes a plurality of frame images; Acquire a first frame image in the first image sequence, perform multi-target vehicle detection on the first frame image through a vehicle detection channel, acquire a multi-target vehicle detection result, and store it in a multi-target vehicle detection result set; Acquire the second frame image ... the Mth frame image in the first image sequence, perform multi-target vehicle detection on the second frame image ... the Mth frame image through a vehicle detection channel, acquire the multi-target vehicle detection result, and store it in the multi-target vehicle detection result set; According to the multi-target vehicle detection result set, the first image sequence is segmented into video streams to obtain the segmented stream data.

4. The method according to claim 3, characterized in that According to the multi-target vehicle detection result, the first image sequence is segmented into video streams to obtain segmented stream data, including: Activate the image segmentation channel, take the multi-target vehicle detection result set as input, and obtain a multi-target vehicle segmentation image set; Performing feature extraction on the multi-target vehicle segmentation image set to obtain a multi-target vehicle feature set, wherein the multi-target vehicle feature set includes vehicle features of K vehicles; A cluster analyzer is constructed to perform cluster analysis based on the multi-target vehicle feature set to obtain the segmented stream data, wherein the segmented stream data includes K groups of segmented stream data sets corresponding to the features of K vehicles in the video stream information.

5. The method according to claim 4, characterized in that Analyze and process the segmented stream data to generate a vehicle position information set, wherein the vehicle position information set includes K groups of vehicle position information, including: Build a vehicle tracker based on object tracking algorithm; Randomly selecting from the segmented stream data to extract a first foreground image sequence; Inputting the first foreground image sequence into the vehicle tracker to obtain first vehicle position information; The remaining foreground image sequences in the segmented stream data are traversed and extracted, the first foreground image sequence is updated, and the vehicle position information set is acquired based on the updated first foreground image sequence.

6. The method according to claim 5, characterized in that Performing correlation evaluation on multiple groups of vehicle position information in the vehicle position information set, obtaining correlation evaluation results, and performing correlation identification of the accident vehicle according to the correlation evaluation results, including: Obtain the location information of the accident vehicle; Traversing the multiple sets of vehicle position information, pairing them with the accident vehicle position information in pairs, and obtaining multiple sets of paired vehicle position information; Performing relative speed analysis on the plurality of groups of paired vehicle position information moment by moment to obtain a relative speed sequence; Performing relative distance analysis on the plurality of groups of paired vehicle position information moment by moment to obtain a relative distance sequence; Based on the correlation evaluation model, correlation evaluation is performed according to the relative speed sequence and the relative distance sequence to obtain the correlation evaluation result.

7. The accident vehicle association identification system based on video stream segmentation is characterized by: The system comprises: A retrieval constraint module, the retrieval constraint module is used to obtain a first accident scene, and generate a video retrieval constraint based on the first accident scene, wherein the video retrieval constraint includes a video time constraint and a video position constraint; A video information acquisition module, wherein the video information acquisition module is used to establish a traffic video information database, wherein the traffic videos in the traffic video information database have position tags and time tags; A retrieval calling module, the retrieval calling module is used to interact with the traffic video information library, perform calling and filtering according to the video retrieval constraint, and obtain filtered video information, wherein the filtered video information includes N video stream information, and N is greater than or equal to 1; A video stream segmentation module, the video stream segmentation module is used to segment the screened video information based on a video stream segmentation algorithm to obtain segmented stream data, wherein the segmented stream data includes K groups of foreground image sequences; A position analysis module, the position analysis module is used to analyze and process the segmented stream data to generate a vehicle position information set, the vehicle position information set includes K groups of vehicle position information; The association evaluation module is used to evaluate the association of multiple groups of vehicle position information in the vehicle position information set, obtain the association evaluation results, and perform association identification of the accident vehicle based on the association evaluation results.

Images