Scene physical information reconstruction method and device and electronic equipment

Abstract

The invention discloses a scene physical information reconstruction method and device and electronic equipment, and relates to the technical field of image processing, and the method comprises the steps: carrying out the target detection of each image frame in to-be-processed target video data, and obtaining the feature information of a target of each image frame; determining a motion track of each target corresponding to the target video data; clustering the motion trail of each target, determining first initial lane data in a target scene corresponding to the target video data, and determining a candidate lane line area corresponding to the target scene; and performing image segmentation processing on the candidate lane line area to obtain second initial lane data, thereby determining a lane line of the target scene. According to the method, the lane line is determined from different processing dimensions of trajectory tracking and image processing, the lane line in the scene does not need to be manually marked, and the problems of low efficiency, high cost and poor scene adaptability of manual scene information marking are solved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method, device and electronic device for reconstructing physical information of a scene. Background Art

[0002] Scene physical information refers to fixed scene information in the video, such as lane lines, stop lines, and zebra crossings. Traffic event detection algorithms in the electric police (electric police is an automatic monitoring device for traffic violations set up on the roadside) all require this scene information, which is currently generally obtained through manual annotation.

[0003] Although manual labeling has high accuracy and strong robustness, it cannot be completed in time when large-scale product deployment occurs, which affects the project schedule. In addition, when the location of equipment such as electric police changes, manual adjustments are required, which is not conducive to later maintenance. It can be seen that manual labeling of scene information has poor adaptability to different scenes, and the labeling cost is high and the efficiency is low. Summary of the invention

[0004] In response to the above problems, the present application provides a method, device and electronic device for reconstructing scene physical information, which solves the problems of low efficiency, high cost and poor scene adaptability of manual scene information labeling.

[0005] In order to achieve the above objectives, this application provides the following technical solutions: A method for reconstructing scene physical information, the method comprising: Performing target detection on each image frame in the target video data to be processed to obtain feature information of the target in each image frame; Determining a motion trajectory of each target corresponding to the target video data based on feature information of the target in each image frame; Clustering the motion trajectory of each target to determine first initial lane data in the target scene corresponding to the target video data; Determining a candidate lane line area corresponding to the target scene based on the first initial lane data; Performing image segmentation processing on the candidate lane line area to obtain second initial lane data; Based on the candidate lane line area and the second initial lane data, a lane line of the target scene is determined.

[0006] Optionally, performing target detection on each image frame in the target video data to be processed to obtain feature information of the target in each image frame includes: Acquire the real-time video stream of the image acquisition device at the target intersection; Performing video decoding on the real-time video stream to obtain target video data, wherein the target video data includes at least one image frame; Based on the target detection model, target detection is performed on each image frame in the target video data to obtain feature information of the target in each image frame, wherein the feature information includes the position of the target in the image frame and the type of the target.

[0007] Optionally, the method further comprises: Obtaining a target data sample set, wherein the target data sample set includes image data of different types of road features, wherein a corresponding detection target is annotated in each image data, and the annotation information corresponding to the detection target at least includes a position and a type of the detection target in the corresponding image data; The target data sample set is processed using an initial detection model so that the initial detection model can output a detection target annotated with each image data in the target data sample set, thereby obtaining a target detection model.

[0008] Optionally, the characteristic information of the target includes a position of the target and a type of the target, and determining the motion trajectory of each target corresponding to the target video data based on the characteristic information of the target of each image frame includes: Based on the type of the target in each image frame, the target is screened to obtain the screened target; Based on the positions of the filtered targets in each image frame, a motion trajectory of each filtered target corresponding to the target video data is determined.

[0009] Optionally, clustering the motion trajectory of each target to determine first initial lane data in the target scene corresponding to the target video data includes: Screening the motion trajectory of each target to obtain the motion trajectory of the screened target; Cluster the motion trajectories of the screened targets and determine the number of initial lanes corresponding to the target scene; Based on the initial lane number, first initial lane data to which the motion trajectory of the filtered target belongs is determined.

[0010] Optionally, determining a candidate lane line area corresponding to the target scene based on the first initial lane data includes: Based on the first initial lane data, determining the position of an average motion trajectory of an object corresponding to each lane; Based on the position of the average motion trajectory of the target corresponding to each lane, a candidate lane line area corresponding to the target scene is determined.

[0011] Optionally, performing image segmentation processing on the candidate lane line area to obtain second initial lane data includes: Performing image segmentation on the candidate lane line area according to an image segmentation model to obtain second initial lane data; The image segmentation model is a neural network model trained based on lane line data samples.

[0012] Optionally, determining the lane line of the target scene based on the candidate lane line area and the second initial lane data includes: Obtaining second position information of a lane line corresponding to the second initial lane data; Determine target position information of each lane line based on the first position information and the second position information of each lane line corresponding to the candidate lane line area; Based on the target position information of each lane line, the lane line of the target scene is determined.

[0013] A scene physical information reconstruction device, the device comprising: A detection unit, configured to perform target detection on each image frame in the target video data to be processed, and obtain feature information of the target in each image frame; A first determining unit, configured to determine a motion trajectory of each target corresponding to the target video data based on feature information of the target in each image frame; A second determining unit, configured to cluster the motion trajectory of each target and determine first initial lane data in a target scene corresponding to the target video data; A third determining unit, configured to determine a candidate lane line area corresponding to the target scene based on the first initial lane data; A segmentation processing unit, used to perform image segmentation processing on the candidate lane line area to obtain second initial lane data; A fourth determination unit is used to determine the lane line of the target scene based on the candidate lane line area and the second initial lane data.

[0014] An electronic device, comprising: A memory, used to store applications and data generated by the operation of the applications; A processor is used to execute the application program to implement the scene physical information reconstruction method as described in any one of the above.

[0015] Compared with the prior art, the present application provides a scene physical information reconstruction method, device and electronic device, the method comprising: performing target detection on each image frame in the target video data to be processed to obtain the characteristic information of the target of each image frame; determining the motion trajectory of each target corresponding to the target video data based on the characteristic information of the target of each image frame; clustering the motion trajectory of each target to determine the first initial lane data in the target scene corresponding to the target video data; determining the candidate lane line area corresponding to the target scene based on the first initial lane data; performing image segmentation processing on the candidate lane line area to obtain the second initial lane data; determining the lane line of the target scene based on the first initial lane data and the second initial lane data. The present application determines the candidate lane line area by the motion trajectory of the target of the target scene, and then the second initial lane data is based on the pixel classification obtained by image processing, so as to determine the lane line from different processing dimensions, without the need to manually mark the lane lines in the scene, thereby solving the problems of low efficiency, high cost and poor scene adaptability of manual scene information marking. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0017] Figure 1 A flow chart of a method for reconstructing physical information of a scene provided in an embodiment of the present application; Figure 2 An electric police viewing angle diagram provided in an embodiment of the present application; Figure 3 A schematic diagram of a lane line modeled based on a motion trajectory provided in an embodiment of the present application; Figure 4 A schematic diagram of a result of deep learning image segmentation provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of a scene physical information reconstruction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] The terms "first" and "second" in this application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may include steps or units that are not listed.

[0020] In an embodiment of the present application, a method for reconstructing scene physical information is provided. The method applies lane line detection to a target scene (such as a monitored road section, a specific location area, etc.). Correspondingly, the detection of stop lines in the target scene is similar to this method. Specifically, the scene physical information includes information such as lane lines, zebra crossings, etc. that are fixed in the camera image. Figure 1 , is a flow chart of a scene physical information reconstruction method provided in an embodiment of the present application, the method may include the following steps: S101 , performing target detection on each image frame in the target video data to be processed, and obtaining feature information of the target in each image frame.

[0021] The target video data to be processed may be pre-collected video data of the corresponding scene to be analyzed, or may be video stream data collected in real time based on an image acquisition device. The target video data includes multiple image frames, each of which corresponds to an image screen. Then, the target in each image frame may be detected according to the target detection algorithm to obtain the characteristic information of the target in each image frame. The characteristic information of the target includes at least the position and type of the target in the image frame. For example, the type may include motor vehicles, non-motor vehicles, and pedestrians. The position of the target represents the position of the target in the current image frame, so that its position in the actual scene road condition can be calculated based on its position in the image frame.

[0022] S102: Determine a motion trajectory of each target corresponding to the target video data based on feature information of the target in each image frame.

[0023] The corresponding target can be screened out according to the target type in each image frame. For example, if the lane line of a motor vehicle needs to be determined, the screened target can be a motor vehicle. Then, the motion trajectory of the target in the target video data is determined according to the position of the screened target in each image frame with a time sequence relationship.

[0024] S103 , clustering the motion trajectory of each target to determine first initial lane data in the target scene corresponding to the target video data.

[0025] After obtaining the motion trajectory of each target, the motion trajectory of the target that meets the conditions can be screened out, such as the motion trajectory of the vehicle that has not changed lanes. Then, cluster analysis is performed on the motion trajectory of the screened targets to calculate the number of lanes in the current scene and the lane to which the motion trajectory of each target belongs, thereby obtaining the first initial lane data, that is, the first initial lane data ensures that the number of lane lines and the position of lane lines in the current scene are obtained based on the motion trajectory analysis of the target.

[0026] S104: Determine a candidate lane line area corresponding to the target scene based on the first initial lane data.

[0027] The first initial lane data includes the position and number of lane lines, etc., and then the approximate lane lines are calculated according to the position of the average trajectory of each lane, and these approximate lane lines are used as candidate lane line areas.

[0028] S105: Perform image segmentation processing on the candidate lane line area to obtain second initial lane data.

[0029] The image processing algorithm in deep learning is used to perform image segmentation processing on the candidate lane line area, that is, to recognize its pixel dimension, so as to obtain the second initial lane data corresponding to the high candidate lane line area, and the second initial lane data represents the information such as the lane line position and number obtained by pixel classification of the candidate lane line area.

[0030] S106: Determine the lane line of the target scene based on the candidate lane line area and the second initial lane data.

[0031] Then, the lane data determined by the motion trajectory of the target is combined with the lane data of the pixel dimension determined based on image processing, so as to obtain a more accurate lane line as the lane line of the target scene.

[0032] The scene physical information reconstruction method provided in the embodiment of the present application extracts the motion trajectory of the target in the scene through the target tracking algorithm, and infers physical information such as lane lines and the number of lanes by modeling the motion trajectory; secondly, the physical information candidate area is determined by the position of the inferred physical information, and then the candidate area is processed using the image segmentation algorithm in deep learning to determine the pixel classification of the candidate area, and finally the physical information of the scene is reconstructed in the image based on the classification results and the results of trajectory modeling. At the same time, this solution can be extended to other scenes with electric police cameras, reducing the cost of manual labeling and increasing the speed of large-scale deployment of products.

[0033] The following describes the relevant technical features in the embodiments of the present application in combination with actual application scenarios.

[0034] In one implementation of an embodiment of the present application, target detection is performed on each image frame in the target video data to be processed to obtain feature information of the target in each image frame, including: acquiring a real-time video stream of an image acquisition device at a target intersection; decoding the real-time video stream to obtain target video data, wherein the target video data includes at least one image frame; and performing target detection on each image frame in the target video data based on a target detection model to obtain feature information of the target in each image frame, wherein the feature information includes the position of the target in the image frame and the type of the target.

[0035] The image acquisition device at the target intersection can be an electric police camera. For example, the rtsp stream address of all electric police cameras at the current intersection is obtained to verify that the rtsp stream address of all electric police cameras at the current intersection is available; secondly, the rtsp stream address of the electric police is accessed to obtain the real-time video of the electric police; then the real-time video is sent to the decoder for video decoding to obtain each image frame in the video, thereby obtaining the target video data including multiple image frames, and subsequent work is based on the analysis of each frame of the image. See Figure 2 , which shows an electric police perspective map provided by an embodiment of the present application, and the electric police perspective map is the image content corresponding to an image frame in the target video data.

[0036] Then, target detection can be performed on each image frame based on the target detection model to obtain target feature information. Specifically, the target detection model can be a pre-trained machine learning model, which can detect different types of targets, such as motor vehicles, pedestrians, non-motor vehicles, etc., and has a high generalization ability. Furthermore, in order to ensure the accuracy of data detection, the image data corresponding to the image frame input to the target detection model needs to be consistent in format with the image data when the target detection model was trained, so as to obtain the location information and type information of the target in each image frame. Specifically, the creation process of the target detection model can include: A target data sample set is obtained, wherein the target data sample set includes image data under different types of road features, wherein a corresponding detection target is annotated in each image data, and the annotation information corresponding to the detection target at least includes the position and type of the detection target in the corresponding image data; the target data sample set is processed using an initial detection model so that the initial detection model can output the detection target annotated with each image data in the target data sample set, and a target detection model is obtained. In the process of the initial detection model learning and processing the target data sample set, it can output the position of the target in the image according to the image, and compare the position of the target output with the actual annotated position corresponding to the target, and adjust the initial detection model according to the deviation data in the comparison result, so as to obtain a target detection model. By detecting the target in the image frame through the target detection model, the efficiency and accuracy of the target detection can be improved. It should be noted that in order to improve the accuracy of the target detection model, the target data sample set for training it needs to include a large amount of data of the expected target, such as selecting to collect a large amount of data under traffic scenes for training, so as to ensure that it has a high generalization ability for each type of intersection. In order to improve the processing efficiency of the target detection model, the image input to the target detection model needs to be preprocessed, that is, the image input to the target detection model cannot be too large, and the decoded image is usually not less than 2 million pixels, so the image needs to be scaled first, and then the image aspect ratio must be kept unchanged during the scaling process.

[0037] In one implementation of the embodiments of the present application, the characteristic information of the target includes the position of the target and the type of the target. Based on the characteristic information of the target in each image frame, the motion trajectory of each target corresponding to the target video data is determined, including: based on the type of the target in each image frame, the targets are screened to obtain screened targets; based on the position of the screened targets in each image frame, the motion trajectory of the target video data corresponding to each screened target is determined.

[0038] First, after obtaining the position information and category information of the target in each image frame, the desired target can be screened, such as only filtering out a specific type of vehicle, which can be determined specifically according to the physical scene information to be reconstructed. Secondly, the trajectory of the screened target can be tracked to obtain the motion trajectory corresponding to the screened target. Specifically, the Kalman filter algorithm can be used to establish a tracking model for the desired target; secondly, the Hungarian algorithm is used to match the target, and the Kalman filter parameter K is updated at the same time; finally, the motion trajectory information of the desired target in the time series is obtained. For example, a set of Kalman filter parameters can be determined according to the target scene to be processed, and then the state covariance matrix and Kalman gain matrix in the Kalman filter algorithm are updated according to the target position information of each image frame. Then it is necessary to determine whether the target in the previous and next frames is the same, and the Hungarian algorithm is used to construct the minimum cost matrix Cost_Matrix, find the most matching target in the previous and next frames for matching, and update the target tracking queue to obtain the motion trajectory of the corresponding target.

[0039] In one implementation of the embodiments of the present application, the motion trajectory of each target is clustered to determine the first initial lane data in the target scene corresponding to the target video data, including: filtering the motion trajectory of each target to obtain the motion trajectory of the filtered target; clustering the motion trajectory of the filtered target to determine the number of initial lanes corresponding to the target scene; based on the number of initial lanes, determining the first initial lane data to which the motion trajectory of the filtered target belongs.

[0040] After obtaining the target's motion trajectory, you can perform trajectory screening and cluster the motion trajectories of targets that meet the conditions. The conditions for motion trajectory screening can be determined based on the actual application scenario, such as only screening the motion trajectories of vehicles that have not changed lanes as the data basis for subsequent cluster analysis. In this way, the number of lane lines and the lane to which each motion trajectory belongs can be determined based on the cluster analysis of the motion trajectory of each target. Figure 3 , which shows a schematic diagram of a lane line modeled based on a motion trajectory provided in an embodiment of the present application. Figure 3 The orange line in the middle is the average trajectory of the lane line determined based on the motion trajectory.

[0041] In one implementation of an embodiment of the present application, based on the first initial lane data, the candidate lane line area corresponding to the target scene is determined, including: based on the first initial lane data, determining the position of the average motion trajectory of the target corresponding to each lane; based on the position of the average motion trajectory of the target corresponding to each lane, determining the candidate lane line area corresponding to the target scene.

[0042] According to the position of the average trajectory of each lane in the candidate lane line area, the approximate lane line position can be calculated, such as Figure 3As shown by the green line in the middle, the area near the green line is used as a candidate lane line area for lane line detection.

[0043] In one implementation of an embodiment of the present application, image segmentation processing is performed on the candidate lane line area to obtain second initial lane data, including: performing image segmentation on the candidate lane line area according to an image segmentation model to obtain second initial lane data; wherein the image segmentation model is a neural network model trained based on lane line data samples.

[0044] The image segmentation model in deep learning can be used to segment the image corresponding to the candidate lane line area to obtain the second initial lane data. Figure 4 , which is a schematic diagram of a result of deep learning image segmentation provided in an embodiment of the present application. The image segmentation model can use the yolov8s-mask lightweight image segmentation network, collect lane line data for training, and use the trained model to detect lane lines on the candidate area.

[0045] In one implementation of the embodiment of the present application, determining the lane line of the target scene based on the candidate lane line area and the second initial lane data includes: Obtain second position information of the lane line corresponding to the second initial lane data; determine the target position information of each lane line based on the first position information and the second position information of each lane line corresponding to the candidate lane line area; determine the lane line of the target scene based on the target position information of each lane line.

[0046] After the candidate lane line area is determined based on the motion trajectory of the target, the first position information corresponding to each lane line in the candidate lane line area can be determined. Then, based on the image segmentation processing, the features of each pixel in the candidate lane line area are determined, and image segmentation is performed based on the pixel features to identify the second position information of each corresponding lane line. If the first position information and the second position information are consistent, the position information is determined as the target position information of the corresponding lane line. If there is a deviation between the two positions, the relevant positions can be optimized according to the deviation range to obtain the lane line of the target scene. That is, the following is obtained. Figure 3 Shown by the blue line.

[0047] In the embodiment of the present application, without adding electric police equipment at the intersection, the video stream is captured and decoded based on the existing electric police, and the vehicle driving trajectory is generated through the vehicle detection and tracking algorithm. The basic scene physical information and the approximate area of ​​the scene physical information are preliminarily calculated through the vehicle trajectory, and then the scene physical information is accurately reconstructed by combining the deep learning method. This method is highly robust and has high adaptability in different scenarios. At the same time, when the camera is affected by the outside world, this method can detect changes in the camera angle in time according to the trajectory of the vehicle, and has strong real-time performance.

[0048] Correspondingly, see Figure 5 , the embodiment of the present application further provides a scene physical information reconstruction device, the device comprising: A detection unit 501 is used to perform target detection on each image frame in the target video data to be processed, and obtain feature information of the target in each image frame; A first determining unit 502 is configured to determine a motion trajectory of each target corresponding to the target video data based on feature information of the target in each image frame; A second determining unit 503 is used to cluster the motion trajectory of each target and determine the first initial lane data in the target scene corresponding to the target video data; A third determining unit 504 is used to determine a candidate lane line area corresponding to the target scene based on the first initial lane data; A segmentation processing unit 505 is used to perform image segmentation processing on the candidate lane line area to obtain second initial lane data; The fourth determining unit 506 is used to determine the lane line of the target scene based on the candidate lane line area and the second initial lane data.

[0049] Optionally, the detection unit includes: A first acquisition subunit is used to acquire a real-time video stream of an image acquisition device at a target intersection; A decoding subunit, configured to perform video decoding on the real-time video stream to obtain target video data, wherein the target video data includes at least one image frame; The detection subunit is used to perform target detection on each image frame in the target video data based on the target detection model to obtain feature information of the target in each image frame, wherein the feature information includes the position of the target in the image frame and the type of the target.

[0050] Optionally, the device further comprises: a model building unit, configured to: Obtaining a target data sample set, wherein the target data sample set includes image data of different types of road features, wherein a corresponding detection target is annotated in each image data, and the annotation information corresponding to the detection target at least includes a position and a type of the detection target in the corresponding image data; The target data sample set is processed using an initial detection model so that the initial detection model can output a detection target annotated with each image data in the target data sample set, thereby obtaining a target detection model.

[0051] Optionally, the characteristic information of the target includes a location of the target and a type of the target, and the first determining unit includes: A first screening subunit, configured to screen the targets based on the type of the targets in each image frame to obtain screened targets; The first determining subunit is used to determine a motion trajectory of each of the filtered targets corresponding to the target video data based on the position of the filtered targets in each image frame.

[0052] Optionally, the second determining unit includes: A second screening subunit is used to screen the motion trajectory of each target to obtain a screened motion trajectory of the target; A clustering subunit is used to cluster the motion trajectories of the screened targets and determine the number of initial lanes corresponding to the target scene; The second determining subunit is used to determine the first initial lane data to which the motion trajectory of the screened target belongs based on the initial lane number.

[0053] Optionally, the third determining unit includes: A third determination subunit, configured to determine a position of an average motion trajectory of an object corresponding to each lane based on the first initial lane data; The fourth determination subunit is used to determine the candidate lane line area corresponding to the target scene based on the position of the average motion trajectory of the target corresponding to each lane.

[0054] Optionally, the segmentation processing unit is configured to: Performing image segmentation on the candidate lane line area according to an image segmentation model to obtain second initial lane data; The image segmentation model is a neural network model trained based on lane line data samples.

[0055] Optionally, the fourth determining unit is configured to: Obtaining second position information of a lane line corresponding to the second initial lane data; Determine target position information of each lane line based on the first position information and the second position information of each lane line corresponding to the candidate lane line area; Based on the target position information of each lane line, the lane line of the target scene is determined.

[0056] Based on the foregoing embodiment, the embodiment of the present application further provides an electronic device, including: A memory, used to store applications and data generated by the operation of the applications; A processor is used to execute the application program to implement the scene physical information reconstruction method as described in any one of the above.

[0057] It should be noted that the above-mentioned processor or CPU can be at least one of an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a Central Processing Unit (CPU), a controller, a microcontroller, and a microprocessor. It is understandable that the electronic device that implements the above-mentioned processor function can also be other, and the embodiments of the present application are not specifically limited. The above is only a specific implementation method of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

[0058] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0059] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for reconstructing scene physical information, characterized in that: The method comprises: Performing target detection on each image frame in the target video data to be processed to obtain feature information of the target in each image frame; Determining a motion trajectory of each target corresponding to the target video data based on feature information of the target in each image frame; Clustering the motion trajectory of each target to determine first initial lane data in the target scene corresponding to the target video data; Determining a candidate lane line area corresponding to the target scene based on the first initial lane data; Performing image segmentation processing on the candidate lane line area to obtain second initial lane data; Based on the candidate lane line area and the second initial lane data, a lane line of the target scene is determined.

2. The method according to claim 1, characterized in that The performing target detection on each image frame in the target video data to be processed to obtain feature information of the target of each image frame includes: Acquire the real-time video stream of the image acquisition device at the target intersection; Performing video decoding on the real-time video stream to obtain target video data, wherein the target video data includes at least one image frame; Based on the target detection model, target detection is performed on each image frame in the target video data to obtain feature information of the target in each image frame, wherein the feature information includes the position of the target in the image frame and the type of the target.

3. The method according to claim 2, characterized in that The method further comprises: Obtaining a target data sample set, wherein the target data sample set includes image data of different types of road features, wherein a corresponding detection target is annotated in each image data, and the annotation information corresponding to the detection target at least includes a position and a type of the detection target in the corresponding image data; The target data sample set is processed using an initial detection model so that the initial detection model can output a detection target annotated with each image data in the target data sample set, thereby obtaining a target detection model.

4. The method according to claim 1, characterized in that: The characteristic information of the target includes the position of the target and the type of the target, and determining the motion trajectory of each target corresponding to the target video data based on the characteristic information of the target of each image frame includes: Based on the type of the target in each image frame, the target is screened to obtain the screened target; Based on the positions of the filtered targets in each image frame, a motion trajectory of each filtered target corresponding to the target video data is determined.

5. The method according to claim 1, characterized in that The clustering of the motion trajectory of each target to determine the first initial lane data in the target scene corresponding to the target video data includes: Screening the motion trajectory of each target to obtain the motion trajectory of the screened target; Clustering the motion trajectories of the screened targets to determine the number of initial lanes corresponding to the target scene; Based on the initial lane number, first initial lane data to which the motion trajectory of the filtered target belongs is determined.

6. The method according to claim 1, characterized in that The determining, based on the first initial lane data, a candidate lane line area corresponding to the target scene includes: Based on the first initial lane data, determining the position of an average motion trajectory of an object corresponding to each lane; Based on the position of the average motion trajectory of the target corresponding to each lane, a candidate lane line area corresponding to the target scene is determined.

7. The method according to claim 1, characterized in that The performing image segmentation processing on the candidate lane line area to obtain second initial lane data includes: Performing image segmentation on the candidate lane line area according to an image segmentation model to obtain second initial lane data; The image segmentation model is a neural network model trained based on lane line data samples.

8. The method according to claim 1, characterized in that: The determining the lane line of the target scene based on the candidate lane line area and the second initial lane data includes: Obtaining second position information of a lane line corresponding to the second initial lane data; Determine target position information of each lane line based on the first position information and the second position information of each lane line corresponding to the candidate lane line area; Based on the target position information of each lane line, the lane line of the target scene is determined.

9. A scene physical information reconstruction device, characterized in that: The device comprises: A detection unit, configured to perform target detection on each image frame in the target video data to be processed, and obtain feature information of the target in each image frame; A first determining unit, configured to determine a motion trajectory of each target corresponding to the target video data based on feature information of the target in each image frame; A second determining unit, configured to cluster the motion trajectory of each target and determine first initial lane data in a target scene corresponding to the target video data; A third determining unit, configured to determine a candidate lane line area corresponding to the target scene based on the first initial lane data; A segmentation processing unit, used to perform image segmentation processing on the candidate lane line area to obtain second initial lane data; A fourth determination unit is used to determine the lane line of the target scene based on the candidate lane line area and the second initial lane data.

10. An electronic device, characterized in that: The electronic device comprises: A memory, used to store applications and data generated by the operation of the applications; A processor is used to execute the application program to implement the scene physical information reconstruction method as described in any one of claims 1 to 8.

Images