An automatic driving vehicle trajectory extraction method and device
By using the KCF algorithm and rotation matrix matching, the problem of limited acquisition range of single-vehicle sensors was solved, enabling continuous tracking and low-cost real-time recording of autonomous vehicle trajectories.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- FAW VOLKSWAGEN AUTOMOTIVE CO LTD
- Filing Date
- 2022-08-30
- Publication Date
- 2026-07-21
AI Technical Summary
In existing autonomous driving technologies, the data coverage of a single vehicle's sensors is limited, making it impossible to learn about a wide range of traffic conditions in a single moment. The collection of obstacle trajectory information is discontinuous and has a high error rate, and the data collection cost is also high.
The KCF algorithm is used to track the region of interest of the target vehicle. By matching the rotation matrix of images at multiple time points and calculating the expanded region, the yaw angle and trajectory of the vehicle are determined. Real-time tracking is achieved by combining the image acquisition module and the matching module.
It improves the success rate of vehicle trajectory tracking, reduces computational costs, solves the target loss problem of traditional algorithms when rotating rapidly, and realizes continuous tracking of large targets.
Smart Images
Figure CN117670924B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of autonomous vehicle trajectory tracking technology, specifically to a method and apparatus for extracting the trajectory of an autonomous vehicle. Background Technology
[0002] In the development of autonomous driving technology, relying on artificial intelligence (AI) is the mainstream approach, requiring substantial datasets. Depending on the specific functional requirements of autonomous driving, various AI algorithms can be employed, each necessitating different dataset designs and varying data requirements. Currently, autonomous driving algorithms heavily utilize deep neural networks to determine obstacle trajectories, thus requiring datasets containing obstacle trajectories for neural network training. However, the current mainstream data acquisition method still relies on onboard sensors to record data, followed by subsequent annotation of obstacle information.
[0003] In the publicly available field of data collection and analysis, autonomous driving data collection relies on sensors installed on the vehicle itself, such as cameras, millimeter-wave radar, and lidar, to collect data in real time as the vehicle travels on the road. The collected data is then used for obstacle trajectory analysis and labeling.
[0004] The current methods for collecting obstacle trajectory information have the following drawbacks: the data coverage of a single vehicle sensor is limited to the area around the vehicle, which is relatively short and cannot learn about traffic conditions over a large distance in a single moment. Obstacles between obstacles can easily cause discontinuous obstacle trajectory information collection and a high error rate. The data collection time of a single vehicle sensor is also relatively high. A single vehicle can only collect data received during its own driving, and the data content is relatively simple. Large-scale data collection by a fleet is needed to expand the types and scenarios of data. Summary of the Invention
[0005] To address at least one aspect of the aforementioned problems, the present invention provides a method for trajectory extraction of an autonomous vehicle, comprising: acquiring a first time-lapse image, the first time-lapse image being a two-dimensional top-down view including a target vehicle; selecting a region of interest (ROI) in the first time-lapse image, the ROI including the target vehicle, the ROI corresponding to a first data set, the first data set including coordinate information and pixel information of the ROI; acquiring a second time-lapse image, the second time-lapse image having the same angle as the first time-lapse image, and the second time-lapse image and the first time-lapse image corresponding to the same region; performing a KCF algorithm with the ROI as the tracking target for calculation, and determining the trajectory in the second time-lapse image based on the calculation result. A candidate region is defined; the candidate region is expanded by a preset factor to obtain an expanded region, the expanded region corresponding to a second data group, the second data group including the coordinate information and pixel information of the expanded region; a rotation angle threshold is set, and the second data group is combined with the two-dimensional rotation matrix corresponding to the rotation angle threshold to calculate an intermediate data group set, each group in the intermediate data group set is template matched with the first data group to determine a third data group; the yaw angle of the target vehicle is determined according to the angle value of the two-dimensional rotation matrix corresponding to the third data group, and the line connecting the center coordinates of the region of interest and the image center coordinates corresponding to the third data group is the trajectory of the target vehicle from the first time moment to the second time moment.
[0006] Preferably, multiple time-lapse images are acquired, and the multiple time-lapse images are sorted according to time. The first image in the sequence of multiple time-lapse images is the first time-lapse image. The multiple time-lapse images have the same angle as the first time-lapse image, and the multiple time-lapse images and the first time-lapse image correspond to the same region. The trajectory of the subsequent time-lapse image relative to the previous time-lapse image is determined and connected according to the time sequence. Then, the trajectory of the target vehicle within the corresponding time period of the multiple time-lapse images is obtained. The region corresponding to the third data group of the subsequent time-lapse image is the region of interest. The trajectory of the target vehicle from the subsequent time-lapse to the next time-lapse is calculated.
[0007] Preferably, the method further includes evaluating the data groups corresponding to the plurality of time-lapse images respectively, and deleting images in the plurality of time-lapse images where the data group is empty.
[0008] Preferably, the rotation angle threshold is 360°, with each degree corresponding to a two-dimensional rotation matrix. The second data set is combined with each two-dimensional rotation matrix to calculate the intermediate data set.
[0009] On the other hand, an autonomous vehicle trajectory extraction device is provided, comprising: an image acquisition module for acquiring a first time-lapse image and a second time-lapse image, wherein the first time-lapse image is a two-dimensional top-down view including the target vehicle, and the second time-lapse image has the same angle as the first time-lapse image and the second time-lapse image corresponds to the same region; and an intermediate calculation module for selecting a region of interest (ROI) in the first time-lapse image, wherein the ROI includes the target vehicle, the ROI corresponds to a first data set, the first data set including the coordinate information and pixel information of the ROI, performing calculations using the KCF algorithm with the ROI as the tracking target, and determining candidate models in the second time-lapse image based on the calculation results. The candidate region is expanded by a preset factor to obtain an expanded region, which includes a second data group containing coordinate and pixel information of the expanded region; a matching module is used to set a rotation angle threshold, and calculate an intermediate data group set by combining the second data group with the two-dimensional rotation matrix corresponding to the rotation angle threshold; each group in the intermediate data group set is template matched with the first data group to determine a third data group; an extraction module is used to determine the yaw angle of the target vehicle based on the angle value of the two-dimensional rotation matrix corresponding to the third data group, and the line connecting the center coordinates of the region of interest and the image center coordinates corresponding to the third data group is the trajectory of the target vehicle from the first time moment to the second time moment.
[0010] Preferably, the image acquisition module acquires multiple time-phase images, which are sorted by time. The first image in the sequence of multiple time-phase images is the first time-phase image. The multiple time-phase images have the same angle as the first time-phase image, and the multiple time-phase images and the first time-phase image correspond to the same region. The extraction module determines and connects the trajectory of the subsequent time-phase image relative to the previous time-phase image in the multiple time-phase images according to the time sequence, thereby obtaining the trajectory of the target vehicle within the corresponding time period of the multiple time-phase images. The region corresponding to the third data group of the subsequent time-phase image is the region of interest, and the trajectory of the target vehicle from the subsequent time-phase to the next time-phase is calculated.
[0011] Preferably, the image acquisition module evaluates the data groups corresponding to the multiple time-time images respectively, and deletes images in the multiple time-time images whose data groups are empty.
[0012] Preferably, the rotation angle threshold of the matching module is 360°, with each degree corresponding to a two-dimensional rotation matrix. The second data group is combined with each two-dimensional rotation matrix to calculate the intermediate data group set.
[0013] The autonomous vehicle trajectory extraction method and apparatus of the present invention have the following beneficial effects: They provide excellent tracking performance for moving obstacles such as vehicles, and are particularly adaptable to situations where vehicles rotate during steering, lane changes, etc. They solve the problem of traditional tracking algorithms losing track of large targets due to rapid rotation changes, leading to significant changes in the characteristics of the original and new targets. This improves the tracking success rate. Furthermore, they address the problem of tracking drift caused by the accumulation of rotation when tracking large targets undergoing continuous small rotation changes, further improving the tracking success rate. The computational cost is low; there is no need for expensive deep learning equipment for time-consuming obstacle recognition data annotation model training. Real-time tracking and recording can begin simply by manually selecting an obstacle once. Attached Figure Description
[0014] To better understand the above and other objects, features, advantages, and functions of the present invention, reference can be made to the embodiments shown in the accompanying drawings. The same reference numerals in the drawings refer to the same parts. Those skilled in the art should understand that the drawings are intended to schematically illustrate preferred embodiments of the invention and do not limit the scope of the invention in any way; the parts in the drawings are not drawn to scale.
[0015] Figure 1 A flowchart illustrating an autonomous vehicle trajectory extraction method according to an embodiment of the present invention is shown;
[0016] Figure 2 A schematic diagram of the region of interest for an autonomous vehicle trajectory extraction method according to an embodiment of the present invention is shown;
[0017] Figure 3 A schematic diagram of candidate regions for an autonomous vehicle trajectory extraction method according to an embodiment of the present invention is shown;
[0018] Figure 4 A schematic diagram of the expanded region rotation of the autonomous vehicle trajectory extraction method according to an embodiment of the present invention is shown;
[0019] Figure 5 A template matching diagram of an autonomous vehicle trajectory extraction method according to an embodiment of the present invention is shown. Detailed Implementation
[0020] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0021] The term "comprising" and its variations as used herein signify open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "at least partially based on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.
[0022] In order to at least partially solve one or more of the above-mentioned problems and other potential problems, one embodiment of this disclosure proposes an autonomous vehicle trajectory extraction method, comprising: acquiring a first moment image, wherein the first moment image is a two-dimensional top-view image including the target vehicle.
[0023] Specifically, an image acquisition device is set up in the area where the target vehicle is located, with the acquisition area defined as a range. The image acquisition device acquires a two-dimensional top-view image of the defined range from a top-down perspective, for example, a view perpendicular to the horizontal plane or a view perpendicular to the plane where the target vehicle is located. Here, the target vehicle is a test vehicle, the defined range is a designated test site, and the acquisition device is fixedly set relative to the test site so that the environmental area covered by its acquired two-dimensional top-down image is always consistent.
[0024] In the first moment image, a region of interest is selected. The region of interest includes the target vehicle. The region of interest corresponds to the first data group, which includes the coordinate information and pixel information of the region of interest.
[0025] like Figure 2 As shown, the region of interest is rectangular, and the entire projection of the target vehicle within the two-dimensional top-view image is included within the region of interest. The coordinates and pixel information of points within the region of interest are obtained through image analysis, where the pixel information includes, but is not limited to, grayscale values. In another embodiment, the shape of the region of interest can also be circular, trapezoidal, rhomboid, or irregular, etc.
[0026] Acquire the second time-phase image. The second time-phase image has the same angle as the first time-phase image, and the second time-phase image and the first time-phase image correspond to the same region.
[0027] Specifically, an image of a second moment within a set range is acquired from the same viewpoint using an image acquisition device. The second moment refers to the moment that is later than the first moment. The interval between the second moment and the first moment can be set according to the calculation requirements, such as 0.1s, 0.5s, 1s, etc.
[0028] The KCF algorithm is used to calculate the region of interest as the tracking target, and the candidate region is determined in the image at the second time step based on the calculation results.
[0029] Specifically, using the region of interest (ROI) as the tracking target, the KCF algorithm is employed for calculation. A global traversal detection is performed on the data containing point coordinates and pixel information corresponding to the second time-time image. Based on the KCF algorithm's response score, the region with the highest probability of similarity to the first data set of the ROI is selected and denoted as the candidate region. Figure 3 As shown.
[0030] The candidate region is expanded by a preset factor to obtain the expanded region. The expanded region corresponds to the second data group, which includes the coordinate information and pixel information of the expanded region.
[0031] Specifically, because the moving target vehicle undergoes displacement and orientation changes between the first and second time points, the tracking area obtained by the original KCF algorithm will drift over time, especially when the vehicle is turning. Therefore, after tracking the region of interest, the candidate region is expanded. For example, the candidate region is expanded by the same number of pixels outside each of its four boundaries to obtain an expanded region, and a second data set including the coordinate information and pixel information of the expanded region is saved.
[0032] Set a rotation angle threshold, and combine the second data group with the two-dimensional rotation matrix corresponding to the rotation angle threshold to calculate the intermediate data group set. Each group in the intermediate data group set is then matched with the first data group to determine the third data group.
[0033] Specifically, the rotation angle threshold is set to ±10°, and the second data set is calculated with the two-dimensional rotation matrix corresponding to the rotation angle threshold at 1° intervals. The formula for the two-dimensional rotation matrix is:
[0034]
[0035] Where θ is the rotation angle, the resulting intermediate data set includes the second data set and 20 intermediate data sets calculated for each rotation angle within the rotation angle threshold. By matching the intermediate data set with the template of the first data set, the intermediate data set with the highest matching score to the first data set is selected, such as... Figure 5 As shown, the record is the third data group.
[0036] The yaw angle of the target vehicle is determined based on the angle value of the two-dimensional rotation matrix corresponding to the third data set. The line connecting the center coordinates of the region of interest and the center coordinates of the image corresponding to the third data set is the trajectory of the target vehicle from the first time moment to the second time moment.
[0037] Specifically, the third data set is calculated from the two-dimensional rotation matrix corresponding to the angles within the rotation angle threshold of the second data set, and the corresponding angle value is the yaw angle of the target vehicle. Based on the image corresponding to the third data set relative to the image of the region of interest, the displacement of the target vehicle between the first and second time moments, i.e., the movement trajectory, can be determined.
[0038] In some embodiments, multiple time-lapse images are acquired and sorted by time. The first image in the sequence of multiple time-lapse images is the first time-lapse image. The multiple time-lapse images have the same angle as the first time-lapse image and correspond to the same region. The trajectory of the subsequent time-lapse image relative to the previous time-lapse image is determined and connected according to the time sequence to obtain the trajectory of the target vehicle within the corresponding time period of the multiple time-lapse images. The region corresponding to the third data group of the subsequent time-lapse image is the region of interest, and the trajectory of the target vehicle from the subsequent time-lapse to the next time-lapse is calculated.
[0039] Specifically, the multiple time-lapse images can be images acquired consecutively within a preset time period, or consecutive frames within a set time period of a video. The multiple time-lapse images are ordered chronologically, for example, first time-lapse image, second time-lapse image, third time-lapse image, ..., N-1 time-lapse image, N time-lapse image. Using the method described above, the trajectory of the target vehicle at the second time relative to the first time-lapse is determined based on the second time-lapse image and the first time-lapse image, the trajectory of the target vehicle at the third time relative to the second time-lapse is determined based on the third time-lapse image and the second time-lapse image, ..., the trajectory of the target vehicle at the N-1 time-lapse is determined based on the N-1 time-lapse image and the N-1 time-lapse image. Connecting these trajectories determines the trajectory of the target vehicle within the preset time period corresponding to the multiple time-lapse images. It should be noted that during the process of extracting the trajectory of the target vehicle at the second time relative to the first time-lapse based on the first time-lapse image and the second time-lapse image, a third data group corresponding to the target vehicle is determined. When extracting the trajectory of the target vehicle from the second time-lapse to the third time-lapse based on the second time-lapse image and the third time-lapse image, the region corresponding to the third data group in the second time-lapse image is the tracking target of the KCF algorithm, i.e., the selected region of interest in the second time-lapse image. By updating the target region being tracked, the problem of tracking result drift or loss caused by target rotation in multiple frames is solved.
[0040] This method is applicable to creating datasets that include obstacle trajectory prediction in autonomous driving, such as one that tracks and records the trajectory of obstacles from a top-down view over a continuous period of time.
[0041] In some embodiments, the method further includes evaluating the data groups corresponding to the multiple time-point images separately and deleting images from the multiple time-point images where the data groups are empty.
[0042] Specifically, when the image acquired by the image acquisition device malfunctions, a blank space will appear, and the data group corresponding to the malfunctioning image will be empty. By evaluating whether the data group corresponding to the image at multiple times is empty, the malfunctioning image can be deleted.
[0043] In some embodiments, the rotation angle threshold is 360°, and each degree corresponds to a two-dimensional rotation matrix. The second data set is combined with each two-dimensional rotation matrix to calculate the intermediate data set.
[0044] Specifically, based on a 360-degree rotation, a two-dimensional rotation matrix is determined at 1-degree intervals. The list of two-dimensional rotation matrices includes 360 matrices. The second data set is then used to calculate the rotation angles of each of these matrices, generating a set of 360 two-dimensional planar rotation data sets. For target vehicles with long time intervals or excessively large rotation angles, possible rotation angles can be covered, improving accuracy.
[0045] On the other hand, an autonomous vehicle trajectory extraction device is provided, comprising: an image acquisition module for acquiring a first-time image and a second-time image, wherein the first-time image is a two-dimensional top-down view including the target vehicle, and the second-time image has the same angle as the first-time image and corresponds to the same region; and an intermediate calculation module for selecting a region of interest (ROI) in the first-time image, the ROI including the target vehicle, the ROI corresponding to a first data set, the first data set including the coordinate information and pixel information of the ROI, performing calculations using the KCF algorithm with the ROI as the tracking target, and determining candidate regions in the second-time image based on the calculation results. The candidate region is expanded by a preset factor to obtain an expanded region, which includes a second data set containing coordinate and pixel information. A matching module sets a rotation angle threshold and combines the second data set with the corresponding two-dimensional rotation matrix to calculate an intermediate data set. Each intermediate data set is then matched with a template from the first data set to determine a third data set. An extraction module determines the yaw angle of the target vehicle based on the angle value of the two-dimensional rotation matrix corresponding to the third data set. The line connecting the center coordinates of the region of interest and the image center coordinates corresponding to the third data set represents the trajectory of the target vehicle from the first time point to the second time point.
[0046] Specifically, the image acquisition module includes a camera positioned above a set range, such as a drone or a road infrastructure camera, to acquire road traffic videos or images with a top-down perspective. The image is then analyzed to determine a data set corresponding to the image, including the coordinates and pixel information of each point. At least part of the functions of one or more of the intermediate calculation module, matching module, and extraction module in the embodiments of this disclosure can be implemented in a single module. Any one or more of the calculation module, matching module, and extraction module according to the embodiments of this disclosure can be implemented by dividing it into multiple modules. One or more of the calculation module, matching module, and extraction module according to the embodiments of this disclosure can be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.
[0047] In some embodiments, the image acquisition module acquires multiple time-series images, which are sorted by time. The first image in the sequence of multiple time-series images is the first time-series image. The multiple time-series images have the same angle as the first time-series image, and the multiple time-series images and the first time-series image correspond to the same region. The extraction module determines the trajectory of the subsequent time-series image relative to the previous time-series image according to the time sequence and connects them to obtain the trajectory of the target vehicle within the corresponding time period of the multiple time-series images. The region corresponding to the third data group of the subsequent time-series image is the region of interest, and the trajectory of the target vehicle from the subsequent time-series to the next time-series is calculated.
[0048] Specifically, the image acquisition module acquires images from multiple time points by connecting to an image acquisition device, and determines the trajectory of the target vehicle contained in the multiple time point images by connecting to the calculation module, matching module, and extraction module. Any one or more of the image acquisition module, calculation module, matching module, and extraction module can be implemented by multiple computer program modules.
[0049] In some embodiments, the image acquisition module evaluates the data groups corresponding to multiple time-time images respectively, and deletes images from the multiple time-time images whose data groups are empty. Specifically, the image acquisition module filters out blank or faulty images by evaluating multiple time-time images.
[0050] In some embodiments, the rotation angle threshold of the matching module is 360°, with each degree corresponding to a two-dimensional rotation matrix. The second data set is combined with each two-dimensional rotation matrix to calculate the intermediate data set.
[0051] Specifically, a two-dimensional rotation matrix is determined at 1° intervals, generating a list of 360 two-dimensional rotation matrices. By pre-storing the list of two-dimensional rotation matrices, the computational power in the trajectory extraction process can be effectively reduced, saving computational costs.
[0052] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand this document.
Claims
1. A method for extracting the trajectory of an autonomous vehicle, characterized in that, include: Acquire a first-moment image, which is a two-dimensional top-view image including the target vehicle; In the first moment image, a region of interest is selected, the region of interest includes the target vehicle, the region of interest corresponds to a first data group, and the first data group includes the coordinate information and pixel information of the region of interest; Acquire a second time-phase image, the second time-phase image has the same angle as the first time-phase image, and the second time-phase image and the first time-phase image correspond to the same region; The KCF algorithm is used to calculate the tracking target of the region of interest, and candidate regions are determined in the image at the second time step based on the calculation results. The candidate region is expanded by a preset multiple to obtain an expanded region. The expanded region corresponds to a second data group, which includes the coordinate information and pixel information of the expanded region. Set a rotation angle threshold, and combine the second data group with the two-dimensional rotation matrix corresponding to the rotation angle threshold to calculate the intermediate data group set. Each group in the intermediate data group set is template matched with the first data group to determine the third data group. The yaw angle of the target vehicle is determined based on the angle value of the two-dimensional rotation matrix corresponding to the third data set. The line connecting the center coordinates of the region of interest and the center coordinates of the image corresponding to the third data set is the trajectory of the target vehicle from the first time moment to the second time moment.
2. The method according to claim 1, characterized in that, Multiple time-series images are acquired and sorted by time. The first image in the sequence of multiple time-series images is the first time-series image. The multiple time-series images have the same angle as the first time-series image and correspond to the same region. The trajectory of the subsequent time-series image relative to the previous time-series image is determined and connected according to the time sequence to obtain the trajectory of the target vehicle within the corresponding time period of the multiple time-series images. The region corresponding to the third data group of the subsequent time-series image is the region of interest. The trajectory of the target vehicle from the subsequent time-series to the next time-series is calculated.
3. The method according to claim 2, characterized in that, It also includes evaluating the data groups corresponding to the multiple time-sharing images separately, and deleting images in the multiple time-sharing images whose data groups are empty.
4. The method according to claim 1, characterized in that, The rotation angle threshold is 360°, and each degree corresponds to a two-dimensional rotation matrix. The second data group is combined with each two-dimensional rotation matrix to calculate the intermediate data group set.
5. An autonomous vehicle trajectory extraction device, characterized in that, include: The image acquisition module is used to acquire a first moment image and a second moment image. The first moment image is a two-dimensional top view image including the target vehicle. The second moment image has the same angle as the first moment image, and the second moment image and the first moment image correspond to the same area. An intermediate calculation module is used to select a region of interest (ROI) in the first time-time image. The ROI includes the target vehicle. The ROI corresponds to a first data group, which includes the coordinate information and pixel information of the ROI. The KCF algorithm is used to perform calculations with the ROI as the tracking target. Based on the calculation results, candidate regions are determined in the second time-time image. The candidate regions are expanded by a preset factor to obtain an expanded region. The expanded region includes a second data group, which includes the coordinate information and pixel information of the expanded region. The matching module is used to set a rotation angle threshold, and calculate an intermediate data set by combining the second data set with the two-dimensional rotation matrix corresponding to the rotation angle threshold. Each group in the intermediate data set is template matched with the first data set to determine the third data set. The extraction module is used to determine the yaw angle of the target vehicle based on the angle value of the two-dimensional rotation matrix corresponding to the third data set. The line connecting the center coordinates of the region of interest and the center coordinates of the image corresponding to the third data set is the trajectory of the target vehicle from the first time moment to the second time moment.
6. The apparatus according to claim 5, characterized in that, The image acquisition module acquires multiple time-series images, which are sorted by time. The first image in the sequence of multiple time-series images is the first time-series image. The multiple time-series images have the same angle as the first time-series image, and the multiple time-series images and the first time-series image correspond to the same region. The extraction module determines and connects the trajectory of the subsequent time-series image relative to the previous time-series image according to the time sequence, thereby obtaining the trajectory of the target vehicle within the corresponding time period of the multiple time-series images. The region corresponding to the third data group of the subsequent time-series image is the region of interest, and the trajectory of the target vehicle from the subsequent time-series to the next time-series is calculated.
7. The apparatus according to claim 6, characterized in that, The image acquisition module evaluates the data groups corresponding to the multiple time-time images respectively, and deletes images in the multiple time-time images whose data groups are empty.
8. The apparatus according to claim 7, characterized in that, The rotation angle threshold of the matching module is 360°, and each degree corresponds to a two-dimensional rotation matrix. The second data group is combined with each two-dimensional rotation matrix to calculate the intermediate data group set.