Trajectory completion, restoration and prediction method, electronic device and readable storage medium

Abstract

The present application discloses a trajectory completion, restoration and prediction method, an electronic device and a readable storage medium, the trajectory completion method comprising: acquiring images collected by multiple acquisition devices arranged in a target area; constructing the original movement trajectory of each moving object based on the collected images; in response to the original movement trajectory being missing, completing the original movement trajectory using a preset trajectory completion rule to obtain the target movement trajectory of the moving object, the preset trajectory completion rule including a plurality of trajectory completion methods arranged in a completion order, thereby improving the accuracy of trajectory completion.

Description

Technical Field

[0001] The present invention relates to the technical field of trajectory processing, and in particular to a trajectory completion method, a trajectory restoration method, a trajectory prediction method, an electronic device and a computer-readable storage medium. Background Art

[0002] In recent years, with the rapid development of artificial intelligence technology, mobile trajectory detection has become one of the hot topics in the field of computer vision research and has been widely used. In the mobile trajectory detection task, it is required to continuously detect the same moving object to obtain the complete trajectory chain of the moving object.

[0003] However, due to problems such as building obstruction, equipment limitations, signal instability, and sampling rate, the trajectory of the moving object determined by the data collected by the collection equipment at each point is sparse or even has missing points, which will directly affect the application effect of subsequent downstream tasks of users.

[0004] Therefore, there is an urgent need for a trajectory completion method to complete the missing parts in the trajectory of a moving object. Summary of the invention

[0005] The main technical problem solved by the present application is to provide a trajectory completion method, a trajectory restoration method, a trajectory prediction method, an electronic device and a computer-readable storage medium, which can improve the accuracy of trajectory completion.

[0006] In order to solve the above technical problems, a technical solution adopted in the present application is: a trajectory completion method is provided, including: acquiring images acquired by multiple acquisition devices arranged in a target area; constructing an original movement trajectory of each moving object based on the acquired images; in response to the existence of missing trajectories in the original movement trajectory, completing the original movement trajectory using preset trajectory completion rules to obtain a target movement trajectory of the mobile object, wherein the preset trajectory completion rules include a plurality of trajectory completion methods arranged in a completion order.

[0007] To solve the above technical problems, another technical solution adopted by the present application is: a trajectory restoration method is provided, including: obtaining a target movement trajectory of a mobile object, wherein the target movement trajectory is based on acquisition images acquired by multiple acquisition devices arranged in a target area; constructing an original movement trajectory of each mobile object based on the acquisition images; in response to the existence of a missing trajectory in the original movement trajectory, completing the original movement trajectory using a preset trajectory completion rule to obtain the target movement trajectory of the mobile object, wherein the preset trajectory completion rule includes a plurality of trajectory completion methods arranged in a completion order; determining a feasible road between adjacent acquisition devices on a navigation map according to the actual installation position of the acquisition device corresponding to each trajectory point in the target movement trajectory; and restoring the target movement trajectory to the navigation map according to the feasible road to obtain a restored trajectory of the mobile object.

[0008] In order to solve the above technical problems, another technical solution adopted by the present application is: a trajectory prediction method is provided, including: obtaining the target movement trajectory of the mobile object, the target movement trajectory is based on acquiring images acquired by multiple acquisition devices set in the target area; constructing the original movement trajectory of each mobile object based on the acquired images; in response to the existence of missing trajectories in the original movement trajectory, using a preset trajectory completion rule to complete the original movement trajectory to obtain the target movement trajectory of the mobile object, the preset trajectory completion rule includes a plurality of trajectory completion methods arranged in a completion order; encoding each trajectory point in the target movement trajectory according to the acquisition time sequence of each trajectory point to obtain each encoded trajectory point; constructing a trajectory feature matrix according to the correlation between each encoded trajectory point; inputting the trajectory feature matrix into a trained trajectory prediction model to obtain the predicted trajectory of the mobile object output by the trained trajectory prediction model.

[0009] To solve the above technical problems, another technical solution adopted in the present application is: to provide an electronic device, including a memory and a processor, the memory storing program instructions, and the processor calling the program instructions from the memory to execute the above-mentioned trajectory completion method and / or trajectory restoration method and / or trajectory prediction method.

[0010] In order to solve the above technical problems, another technical solution adopted in the present application is: providing a computer-readable storage medium including program data stored therein, wherein the program data is used to implement the above trajectory completion method and / or trajectory restoration method and / or trajectory prediction method when executed by a processor.

[0011] Compared with the traditional method of using a single completion method to complete the missing trajectory, the above scheme is prone to incomplete completion. The present application proposes a trajectory completion method, which includes: obtaining acquisition images acquired by multiple acquisition devices set in the target area; constructing the original movement trajectory of each moving object based on the acquisition image; in response to the original movement trajectory having a missing trajectory, using a preset trajectory completion rule to complete the original movement trajectory to obtain the target movement trajectory of the moving object, and the preset trajectory completion rule includes a variety of trajectory completion methods arranged in order of completion. Therefore, the original movement trajectory with missing trajectories is completed by using a variety of trajectory completion methods in order of priority, and the completion possibility is deeply explored from different data angles, thereby improving the completion accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] 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 are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0013] Figure 1 is a flowchart of an exemplary embodiment of a trajectory completion method shown in the present application;

[0014] Figure 2 is a schematic diagram of an exemplary embodiment of an original movement trajectory shown in the present application;

[0015] Figure 3 yes Figure 1 A schematic flow chart of an exemplary embodiment of step S130 in the trajectory completion method is shown;

[0016] Figure 4 is a flow chart of an exemplary embodiment of a trajectory restoration method shown in the present application;

[0017] Figure 5 is a schematic diagram of an exemplary embodiment of a restoration trajectory shown in the present application;

[0018] Figure 6 is a flowchart of an exemplary embodiment of a trajectory prediction method shown in the present application;

[0019] Figure 7 It is a schematic diagram of the framework of the trajectory completion, restoration and prediction method shown in this application;

[0020] Figure 8 It is a specific framework schematic diagram of the trajectory completion, restoration and prediction method shown in this application;

[0021] Fig. 9 is a structural schematic diagram of an exemplary embodiment of a trajectory completion device shown in the present application;

[0022] Fig.10 It is a structural schematic diagram of an embodiment of an electronic device provided by the present application;

[0023] Fig.11 It is a structural diagram of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It will be appreciated that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some but not all structures related to the present application are shown in the drawings. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the art without making creative work are within the scope of protection of the present application.

[0025] First of all, it should be noted that the prediction and restoration of movement trajectories of large amounts of traffic travel data based on machine learning and artificial intelligence methods can not only provide a basis for analyzing the activity patterns between mobile objects and roads in a certain area, but is also crucial for judging the associated places of mobile objects and mining accompanying relationships.

[0026] Traditional trajectory completion methods are mostly based on probability models, nearest neighbor methods, etc., which cannot effectively obtain nonlinear relationships in trajectory data, resulting in incomplete and inaccurate trajectory completion. For example, the trajectory completion method based on the interpolation principle uses adjacent sample data for simple linear interpolation, but the distribution of roads and collection points does not conform to linearity, which will directly lead to errors in interpolation points. The typical manifestation is that the added trajectory points are scattered outside the road.

[0027] Based on this, the present application proposes a trajectory completion method, a trajectory restoration method, a trajectory prediction method, an electronic device and a computer-readable storage medium, which can improve the accuracy of trajectory completion. Figure 1 , Figure 1 It is a flowchart of an exemplary embodiment of a trajectory completion method shown in the present application.

[0028] The execution subject of the trajectory completion method may be a terminal device or a server or other processing device, wherein the terminal device may be a user equipment (UE), a computer, a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. The execution subject of the trajectory completion method may also be a trajectory completion device. In some possible implementations, the trajectory completion method may be implemented by a processor calling a computer-readable instruction stored in a memory.

[0029] Specifically, the trajectory completion method of this embodiment includes the following steps:

[0030] S110: Acquire captured images captured by multiple capturing devices arranged in the target area.

[0031] The target area may be a pre-divided area, or a city, etc. In practical applications, the target area may be divided according to the moving speed and moving time of the moving object.

[0032] The acquisition device can be used to acquire images including moving objects. Exemplarily, the acquisition device can be an image acquisition device, a video acquisition device, etc.

[0033] The captured image refers to an image to be identified. Exemplarily, the captured image can be acquired by a capture device. In addition, an image whose image quality is greater than a preset quality threshold among multiple images captured by the capture device can also be used as the captured image.

[0034] The trajectory completion device determines the target area and obtains the collected images of multiple collection devices in the target area within a preset time period. For example, the collected images collected by the collection devices set at each collection point in a city within the preset time period can be obtained.

[0035] S120: constructing an original moving trajectory of each moving object based on the collected images.

[0036] A moving object may be something with movable features in a captured image. For example, a moving object may be an animal, a vehicle, a train, etc. in a captured image. A captured image may contain multiple moving objects, and a moving object may also exist in multiple captured images. Object detection processing is performed on the multiple captured images to obtain the moving objects included in the multiple captured images.

[0037] The original moving trajectory refers to the path or route formed by the moving object when it moves. Exemplarily, feature extraction can be performed on the image captured by the acquisition device to determine the moving object captured by the acquisition device, and the moving objects captured by each acquisition device and the corresponding time can be recorded; the points of the acquisition devices are connected according to the appearance time of the same moving object in each acquisition device to obtain the original moving trajectory of the corresponding moving object. In other embodiments, video frames captured by multiple acquisition devices containing mobile objects are obtained, and the time points of the video frames of the mobile objects captured by the multiple acquisition devices are determined, and the video frames about the mobile objects are merged according to the order of the time points to obtain the original moving trajectory of the mobile object. In other embodiments, the GPS position of the mobile object can be combined with the acquisition image captured by the acquisition device to obtain the original moving trajectory of each mobile object.

[0038] For details, please refer to Figure 2 , take the area shown in the box in the figure as the target area, and the dots marked with A, B, C, etc. are the acquisition devices that have acquired the same moving object. Connect the acquisition devices of AK in sequence to obtain the original moving trajectory of the moving object (shown by the solid line in the figure).

[0039] S130: In response to the original moving trajectory having missing tracks, a preset track completion rule is used to complete the original moving trajectory to obtain a target moving trajectory of the moving object, wherein the preset track completion rule includes a plurality of track completion methods arranged in a completion order.

[0040] Trajectory loss refers to the presence of undetected trajectory points in the original moving trajectory. Exemplarily, a trajectory loss detection method can be used to detect whether the original moving trajectory of a moving object is missing. For example, the consistency check of the moving object's motion can be used. If the moving speed or direction of the moving object suddenly changes, resulting in discontinuous movement, it indicates that the original moving trajectory may be missing; background modeling or background subtraction technology can also be used. If no background changes related to the moving object are detected in some frames, it indicates that the original moving trajectory of the moving object may be missing or wrong.

[0041] The preset trajectory completion rule refers to completing the missing original moving trajectory in the order of completion of multiple trajectory completion methods. The multiple trajectory completion methods in the preset trajectory completion rule are different trajectory completion methods. By using different trajectory completion methods to complete the missing original moving trajectory, the target moving trajectory of the moving object is obtained. Exemplarily, the trajectory completion methods in the preset trajectory completion rule can be 2, 3, 4 or 5, etc. Of course, it can also be the same trajectory completion method with different parameters.

[0042] The trajectory completion method is used to complete the missing points in the trajectory. Exemplarily, the trajectory completion method can be spatiotemporal analysis waste film recall, probability-based individual and global completion algorithm, different moving object inspection based on fusion data, license plate anomaly detection completion algorithm, trajectory completion based on adjoint algorithm, and completion algorithm based on trajectory similarity, etc. The preset trajectory completion rules may include the above two or more trajectory completion methods.

[0043] The target moving trajectory refers to the completed moving trajectory. Exemplarily, the original moving trajectory with missing trajectories is completed in sequence according to the completion order by using multiple trajectory completion methods to obtain the target moving trajectory of the moving object.

[0044] It can be seen that the trajectory completion method of the embodiment of the present application obtains the captured images captured by multiple capture devices set in the target area; constructs the original movement trajectory of each moving object based on the captured images; in response to the original movement trajectory having missing trajectories, the original movement trajectory is completed using the preset trajectory completion rules to obtain the target movement trajectory of the moving object, and the preset trajectory completion rules include a variety of trajectory completion methods arranged in the order of completion. Therefore, the original movement trajectory with missing trajectories is completed using a variety of trajectory completion methods in order of priority, and the completion possibility is deeply explored from different data angles, thereby improving the completion accuracy.

[0045] Based on the above embodiments, the embodiments of the present application adopt Figure 3 The flowchart details how to complete the original moving trajectory with missing trajectories according to the preset trajectory completion rules. Figure 3 , Figure 3 yes Figure 1 The flowchart of an exemplary embodiment of step S130 in the trajectory completion method is shown. Specifically, step S130 uses a preset trajectory completion rule to complete the original movement trajectory, and the process of obtaining the target movement trajectory of the moving object specifically includes the following steps:

[0046] S310: performing a first processing on the original movement trajectory based on the first trajectory completion method to obtain a processed movement trajectory.

[0047] First, in order to distinguish different moving objects, the moving object corresponding to the moving trajectory to be completed can be used as the target object. That is, the original moving trajectory of the target object has a missing trajectory and needs to be completed.

[0048] In some embodiments, the first trajectory completion method can be a spatiotemporal analysis waste film recall. Wherein, waste films refer to images that have not been successfully matched, that is, images that have not been matched with moving objects. Spatiotemporal analysis waste film recall includes a time period high-frequency acquisition waste film recall and a spatial history transfer law recall algorithm.

[0049] Recall of waste films collected at high frequency in a time period refers to analyzing the historical collection data of the target object, obtaining the points where the target object appears at high frequency in a certain period of time, and recalling and reanalyzing the images collected at the corresponding points in the corresponding period of time. Points refer to the collection positions of each collection device. Specifically, 24 hours a day is divided into multiple time periods, and the correspondence between the time periods and the points is statistically analyzed in combination with the historical collection trajectory of the target object to obtain the high-frequency time periods and the corresponding high-frequency points of the target object. For example, the target object often appears at location 1 from 7 to 9 in the morning. If there is a lack of high-frequency points in the original moving trajectory of the target object, mobile object detection is performed on the collected images collected at the missing high-frequency points in the corresponding high-frequency time period, and features are extracted for each detected moving object to obtain object features of each moving object. For example, if the trajectory points of the target object at location 1 from 7 to 9 in the morning are missing in the original moving trajectory, the collected images collected by the collection device at location 1 from 7 to 9 in the morning are re-analyzed; the object features of each moving object are compared with the target object features of the target object to obtain feature similarity; if the feature similarity is greater than or equal to the feature similarity threshold, the corresponding moving object is characterized as the target object, the corresponding point is used as the missing point in the original moving trajectory, and the specific time when the target object is detected is recorded; if the feature similarity is less than the feature similarity threshold, the corresponding moving object is characterized as not the target object. Among them, the feature comparison methods include average centroid similarity and same-day collection similarity. Average centroid similarity is to compare the features of each mobile object with the historical collection images of the target object, and calculate the average value to obtain the average centroid similarity of each mobile object and the target object; same-day collection similarity is to compare the features of each mobile object with the same-day collection images of the target object to obtain the same-day collection similarity of each mobile object and the target object. Average centroid similarity can obtain the features of the target object at different angles for comparison, and same-day collection similarity can obtain the features of the target object on that day, such as clothing. Average centroid similarity can be determined as the feature similarity of each mobile object and the target object; same-day collection similarity can also be determined as the feature similarity of each mobile object and the target object; average centroid similarity and same-day collection similarity can also be weighted averaged to obtain the feature similarity of each mobile object and the target object.

[0050] The spatial historical transfer law recall algorithm refers to analyzing the historical acquisition data of the target object, obtaining the transfer law of the target object between each point, and determining the point of waste film recall based on the transfer law. Specifically, the historical movement trajectory of the target object is analyzed to determine the regular transfer path of the target object. For example, if the regular transfer path of the target object is from A to B and then to C, a dictionary {B:{past:[A], next:[C]}} can be generated, that is, if there is a trajectory point B in the original trajectory of the target object, it can be determined based on the spatial historical transfer law of the target object that the previous trajectory point of the target object may be A, and the next trajectory point may be C. The acquisition images collected by the acquisition devices corresponding to points A and C in the corresponding time period can be re-analyzed to determine whether points A and C are missing trajectory points.

[0051] In other embodiments, the first trajectory completion method may also be a probability-based completion algorithm, a different moving object inspection based on fused data, a license plate anomaly detection completion algorithm, a trajectory completion based on an adjoint algorithm, and a trajectory similarity-based completion algorithm.

[0052] The probability-based completion algorithm specifically includes the following steps: clustering the acquisition devices with adjacent distances, for example, clustering the acquisition devices within one meter or two meters into one category, thereby reducing the problems of short completion distance and repeated completion; determining the individual probability transfer matrix of the target object through the historical movement trajectory of the target object, that is, the probability of the target object moving from one point to another. For example, through the historical movement trajectory of the target object, the target object moves from A to B and then to C 11 times, and moves directly from A to C 3 times, then the probability of the target object moving from A to B is 11 / 14, and the probability of moving from A to C is 3 / 14, thereby obtaining the individual probability transfer matrix of the target object; completing the original movement trajectory of the target object through the individual probability transfer matrix. In other application scenarios, the global probability transfer matrix is ​​determined through the transfer probability of all mobile objects from one point to another, and the original movement trajectory of the target object is completed through the global probability transfer matrix.

[0053] The completion algorithm based on trajectory similarity includes a strong adjoint algorithm module and a trajectory similarity calculation algorithm based on edit distance. The strong adjoint algorithm module specifically includes the following steps: analyzing the captured image including the target object, if a moving object and the target object are captured by different acquisition devices or the same acquisition device at different time points, it indicates that there is a companion relationship between the moving object and the target object; the original moving trajectory of the target object can be completed through the moving trajectory of the moving object.

[0054] The inspection of different moving objects based on fused data is usually the inspection of the association relationship of vehicles, and the subordinate external data sources of different moving objects can be called. Specifically, the following steps are included: if the vehicle and the target object are collected by different collection devices or the same collection device at different time points, it indicates that there is an association relationship between the vehicle and the target object; the movement trajectory of the vehicle can be obtained, and the original movement trajectory of the target object can be completed through the movement trajectory of the vehicle.

[0055] Taking the recall of waste films through spatiotemporal analysis as an example, the trajectory completion device determines the points of waste film recall through the high-frequency collection of waste film recall in time periods and the recall algorithm of spatial historical transfer rules, and compares the features of each moving object in the waste films of the corresponding points and corresponding time periods with the target objects to obtain the missing points, and completes the original moving trajectory through the missing points to obtain the processed moving trajectory.

[0056] In other embodiments, the trajectory completion device can also complete the original moving trajectory based on the first trajectory completion method to obtain a completed moving trajectory; verify the added trajectory points in the completed moving trajectory to obtain a verification result; in response to the verification result indicating that there are no trajectory points with completion errors in the added trajectory points, the completed moving trajectory is used as the processed moving trajectory; in response to the verification result indicating that there are trajectory points with completion errors in the added trajectory points, the trajectory points with completion errors are removed from the completed moving trajectory, and the moving trajectory after removing the error trajectory points is used as the processed moving trajectory. In this way, the trajectory points added by the first trajectory completion method are verified to avoid adding error trajectory points and ensure the accuracy of trajectory completion. At the same time, the verification process based on trajectory completion will analyze the video stream at specific points and specific time periods, which is highly directional and efficient.

[0057] After each trajectory completion method completes the trajectory, the trajectory points added by each trajectory completion method can be verified to verify whether the added trajectory points are correct trajectory points. The added trajectory points refer to the trajectory points added to the moving trajectory completed by each trajectory completion method relative to the moving trajectory completed by the previous trajectory completion method, or can be the trajectory points added to the moving trajectory completed by the first trajectory completion method relative to the original moving trajectory. Exemplarily, the original moving trajectory is completed by the first trajectory completion method to obtain the trajectory points added by the first trajectory completion method, and the trajectory points added by the first trajectory completion method are verified, the moving trajectory processed by the first trajectory completion method is completed by the second trajectory completion method to obtain the trajectory points added by the second trajectory completion method, and the trajectory points added by the second trajectory completion method are verified, and so on. Exemplarily, the number of trajectory points added by the first trajectory completion method may be multiple. If some of the added trajectory points are erroneous trajectory points, the erroneous trajectory points are removed from the completed moving trajectory, and the original moving trajectory and the correctly completed trajectory points are connected to obtain the moving trajectory processed by the first trajectory completion method.

[0058] As an example, the steps of verifying the added trajectory points include: taking the moving object corresponding to the completed moving trajectory as the target object; obtaining the object features of the moving object in the video frame collected by the collection device corresponding to each added trajectory point within the collection time period, the collection time period representing the collection time of the target object collected by the collection device; calculating the first feature similarity between the object feature of the moving object and the target object feature of the target object; in response to the first feature similarity being greater than or equal to the first preset feature similarity threshold, the added trajectory point is characterized as a correctly completed trajectory point; in response to the first feature similarity being less than the first preset feature similarity threshold, the added trajectory point is characterized as an incorrectly completed trajectory point.

[0059] In order to distinguish different moving objects, the moving object corresponding to the completed moving trajectory can be used as the target object. That is, the original moving trajectory of the target object has a missing trajectory and needs to be completed.

[0060] When determining the trajectory point of the target object, the corresponding position of the acquisition device that acquires the target object can be determined as the trajectory point of the target object, and the various trajectory points of the target object are connected according to the acquisition time of the trajectory point to obtain the moving trajectory of the target object. Therefore, in order to verify whether the trajectory point added by the trajectory completion method is correct, it can be determined whether the target object appears in the video stream collected by the acquisition device corresponding to the added trajectory point within the acquisition time period. If it appears, the verification is successful, and if it does not appear, the verification fails. Among them, the acquisition time period refers to the acquisition time of the target object collected by the acquisition device corresponding to the added trajectory point. Exemplarily, in the first trajectory completion method, if a certain acquisition device is determined to be the trajectory point of the target object, the specific time period for the acquisition device to collect the target object will also be recorded, and the specific time period can be used as the acquisition time period in the verification process. For example, the location 1 corresponding to the target object is used as the added trajectory point, and the specific time period for the acquisition device at location 1 to collect the target object is 8 am-8.05 am, then 8 am to 8.05 am is used as the acquisition time period, and the video stream collected by the acquisition device at location 1 during this time period is obtained.

[0061] After obtaining the video stream collected by the collection device corresponding to the added trajectory point within the collection time period, target detection is performed on each video frame in the video stream to obtain the moving object in each video frame, feature extraction is performed on the moving object in each video frame to obtain the object feature of the moving object in each video frame, and the first feature similarity between the object feature of the moving object in each video frame and the target object feature of the target object is calculated; if the first feature similarity is greater than or equal to the first preset feature similarity threshold, it indicates that the moving object in the video stream is the target object, that is, the corresponding added trajectory point is the correct trajectory point for completion; if the first feature similarity is less than the first preset feature similarity threshold, it indicates that each moving object in the video stream is not the target object, that is, the corresponding added trajectory point is the wrong trajectory point for completion. Exemplarily, the calculation method of the first feature similarity can be cosine similarity, Euclidean distance, etc. The target object feature of the target object can be extracted from the ID photo of the target object, or from the historical collection image including the target object. It is not specifically limited here, as long as it is an image including the target object.

[0062] As another example, the step of verifying the added trajectory points also includes: obtaining the object features of the mobile object in the video frame collected by the collection device corresponding to each added trajectory point within the collection time period, the collection time period represents the collection time of the collection device collecting the target object, and the target object is the mobile object corresponding to the completed mobile trajectory; determining the similarity scaling parameters of each mobile object according to the spatiotemporal distance between each mobile object and other trajectory points in the video frame, and the other trajectory points represent the trajectory points other than the added trajectory points in the original mobile trajectory; determining the product between the similarity scaling parameters of each mobile object and the corresponding first feature similarity as the second feature similarity of the corresponding mobile object, and the first feature similarity represents the feature similarity between the object features of each mobile object and the target object features of the target object; in response to the second feature similarity being greater than or equal to the first preset feature similarity threshold, the added trajectory point is represented as a correctly completed trajectory point; in response to the second feature similarity being less than the first preset feature similarity threshold, the added trajectory point is represented as an incorrectly completed trajectory point. Thus, scaling the first feature similarity by the similarity scaling parameter can increase the verification accuracy, thereby improving the adaptability of the verification process to the erroneous trajectory points.

[0063] Among them, other trajectory points can be the trajectory points with the closest spatial distance to the corresponding added trajectory points, or can be the trajectory points with the closest temporal distance. The temporal and spatial representations of other trajectory points correspond to the setting position of the acquisition device and the time of acquiring the target object. The video stream acquired by the acquisition device corresponding to the added trajectory point during the acquisition time period includes multiple video frames, each video frame may include multiple moving objects, and the position of a moving object in multiple video frames may also be different. Any trajectory point other than the added trajectory point in the original moving trajectory is used as a reference for other trajectory points, and the time and space of each moving object in different video frames and the time and space of other trajectory points are calculated to obtain the time and space distance between the two. Among them, the time and space distance may include time distance and space distance, or may only include time distance or only include space distance. When calculating the time and space distance, the position and corresponding acquisition time of the moving object in each video frame can be analyzed, and the position and corresponding acquisition time of the target object in the acquisition image including the target object can be analyzed by other trajectory points, thereby determining the time and space distance according to the position and corresponding acquisition time of the moving object in each video frame and the position and corresponding acquisition time of the target object.

[0064] The similarity scaling factor is used to scale the first feature similarity between the object feature of each moving object and the target object feature of the target object. Exemplarily, after obtaining the spatiotemporal distance between each moving object and other trajectory points in the video frame, the similarity scaling factor is determined according to the spatiotemporal distance, and the similarity scaling factor has spatiotemporal attenuation. The range of the similarity scaling factor can be 0-1. Exemplarily, the similarity scaling factor can be determined according to the time distance, and the longer the time distance is, the smaller the similarity scaling factor is, and thus the product between the first feature similarity and the similarity scaling factor is also smaller; the shorter the time distance is, the larger the similarity scaling factor is, and thus the product between the first feature similarity and the similarity scaling factor is also larger. The similarity scaling factor can also be determined according to the spatial distance, and the longer the spatial distance is, the smaller the similarity scaling factor is, and thus the product between the first feature similarity and the similarity scaling factor is also smaller; the shorter the spatial distance is, the larger the similarity scaling factor is, and thus the product between the first feature similarity and the similarity scaling factor is also larger. The above-mentioned time distance and space distance can also be combined by weighting to obtain a similarity scaling factor. In addition, generally, the longer the time is, the longer the moving distance is, so the similarity scaling factor can also be determined based on the time distance and the space distance. The longer the time distance is and the shorter the space distance is, the smaller the similarity scaling factor is, and thus the product between the first feature similarity and the similarity scaling factor is also smaller; the shorter the time distance is and the longer the space distance is, the smaller the similarity scaling factor is, and thus the product between the first feature similarity and the similarity scaling factor is also smaller; and the shorter the time distance is and the shorter the space distance is, or the longer the time distance is and the longer the space distance is, the larger the similarity scaling factor is, and thus the product between the first feature similarity and the similarity scaling factor is also larger.

[0065] S320: performing a second processing on the processed movement trajectory based on the second trajectory completion method to obtain a target movement trajectory of the moving object.

[0066] The second trajectory completion method can be spatiotemporal analysis waste film recall, probability-based completion algorithm, different moving object inspection based on fusion data, license plate anomaly detection completion algorithm, trajectory completion based on adjoint algorithm, and completion algorithm based on trajectory similarity, etc. It should be noted that in order to deeply explore the completion possibilities from different angles, the second trajectory completion method and the first trajectory completion method can choose different trajectory completion methods, for example, the first trajectory completion method is spatiotemporal analysis waste film recall, and the second trajectory completion method is a probability-based completion algorithm.

[0067] In other embodiments, the preset trajectory completion rules include a first trajectory completion method, a second trajectory completion method, and a third trajectory completion method arranged in order of completion. The trajectory completion device performs a first processing on the original moving trajectory based on the first trajectory completion method to obtain the moving trajectory after the first processing; performs a second processing on the moving trajectory after the first processing based on the second trajectory completion method to obtain the moving trajectory after the second processing; performs a third processing on the moving trajectory after the second processing based on the third trajectory completion method to obtain the target moving trajectory of the moving object. As an example, the first trajectory completion method is spatiotemporal analysis waste film recall, the second trajectory completion method is a probability-based completion algorithm, and the third trajectory completion method is a trajectory similarity-based completion algorithm.

[0068] In other embodiments, the preset trajectory completion rules include a first trajectory completion method, a second trajectory completion method, a third trajectory completion method, and a fourth trajectory completion method arranged in order of completion. The trajectory completion device performs a first processing on the original moving trajectory based on the first trajectory completion method to obtain the moving trajectory after the first processing; performs a second processing on the moving trajectory after the first processing based on the second trajectory completion method to obtain the moving trajectory after the second processing; performs a third processing on the moving trajectory after the second processing based on the third trajectory completion method to obtain the moving trajectory after the third processing; performs a fourth processing on the moving trajectory after the third processing based on the fourth trajectory completion method to obtain the target moving trajectory of the moving object. As an example, the first trajectory completion method is spatiotemporal analysis waste film recall, the second trajectory completion method is a probability-based completion algorithm, the third trajectory completion method is a trajectory similarity-based completion algorithm, and the fourth trajectory completion method is a different moving object inspection based on fused data.

[0069] It can be seen that the trajectory completion method of the embodiment of the present application performs serial trajectory completion through the first trajectory completion mode and the second trajectory completion mode, and multiple trajectory completion modes are used in combination to explore completion possibilities from different angles, thereby improving the accuracy of trajectory completion.

[0070] Based on the above embodiments, the embodiments of the present application adopt Figure 4 The flowchart details how to restore the target movement trajectory to the navigation map. Figure 4 , Figure 4 FIG. 1 is a flow chart of an exemplary embodiment of a trajectory restoration method shown in the present application. Specifically, the following steps are included:

[0071] S410: Acquire the target moving trajectory of the moving object.

[0072] The target moving trajectory is obtained by the above-mentioned trajectory completion method, and the main acquisition method is based on acquiring the acquisition images acquired by multiple acquisition devices set in the target area; constructing the original moving trajectory of each moving object based on the acquisition images; in response to the missing trajectory of the original moving trajectory, the original moving trajectory is completed using the preset trajectory completion rules to obtain the target moving trajectory of the moving object. The preset trajectory completion rules include a variety of trajectory completion methods arranged in order of completion. The target moving trajectory obtained by the trajectory completion method belongs to the display of trajectory points. The trajectory restoration method combines the trajectory points in the target moving trajectory and the map to connect the trajectory points with the road to form a complete path.

[0073] The trajectory restoration device obtains the target moving trajectory of the moving object completed by the trajectory completion device.

[0074] S420: Determine feasible roads between adjacent collection devices from the navigation map according to actual installation positions of collection devices corresponding to each track point in the target moving track.

[0075] Each track point in the target moving track may be the actual installation location of the collection device of the mobile object. The actual installation location refers to the geographical location of the collection device in the navigation map. Exemplarily, the geographical location of the collection device may be obtained through a satellite positioning system. Among them, geographical features such as roads, buildings, and landmarks are displayed in the navigation map. The satellite positioning system may be a GPS, GNSS system, etc.

[0076] Adjacent collection devices may be collection devices whose collection times of the mobile objects are adjacent, or collection devices that are adjacent in space. Figure 5 , the collection devices are sorted from earliest to latest according to the collection time of the mobile object, and the collection device AN is obtained, that is, the collection time of the collection device A is the earliest, and the collection time of the collection device N is the latest. It can be understood that the mobile object moves from A to N in alphabetical order. Among them, the collection device A and the collection device B are adjacent collection devices, the collection device B and the collection device C are adjacent collection devices, and so on.

[0077] A feasible road refers to a passable road for a mobile object to move from one collection device to another collection device among adjacent collection devices. The passable road may be a road existing in the navigation map.

[0078] Exemplarily, in response to the actual installation position of the acquisition device being located on the road, the trajectory restoration device determines the road corresponding to the acquisition device as a feasible road; in response to the actual installation position of the acquisition device not being located on the road, the adjacent road of the acquisition device is determined as a feasible road according to the trajectory end point of the target moving trajectory; the feasible roads of adjacent acquisition devices are connected at the intersection between the feasible roads of adjacent acquisition devices to obtain the feasible roads between adjacent acquisition devices. Thus, feasible roads are determined by different actual installation positions of the acquisition devices, and when the actual installation position of the acquisition device is not located on the road, the adjacent road is determined as a feasible road, which improves the adaptability and accuracy of trajectory restoration and reduces the computational load of feasible path search.

[0079] When the actual installation location of the acquisition device is located on the road, it is also divided into being installed at the end of the road or in the middle of the road. When the acquisition device is installed at the end of the road, the road can be directly used as a feasible road; when the acquisition device is installed in the middle of the road, the road can be divided into two roads according to the acquisition device and used as feasible roads. When the actual installation location of the acquisition device is not located on the road, a common method is to perform point adsorption or point mapping, and adsorb the acquisition device to the adjacent road according to the trajectory end point of the target moving trajectory. Exemplarily, point adsorption may include a nearest neighbor method or a distance-based adsorption method. The adjacent road in the direction of the trajectory end point of the target moving trajectory of the acquisition device is used as a feasible road. For example, if the trajectory end point direction of the target moving trajectory is to the south of the acquisition device, the adjacent road in the south direction of the acquisition device is selected as a feasible road. The adjacent road of the acquisition device can also be determined according to the direction of the adjacent acquisition device of the acquisition device. Specifically, the acquisition device can be adsorbed to the adjacent road according to the actual situation. In the above manner, a graph with the acquisition device as the vertex and the feasible road corresponding to the acquisition device as the edge can be formed. Among them, one acquisition device may correspond to multiple feasible roads.

[0080] S430: Restoring the target moving trajectory onto the navigation map according to the feasible roads to obtain the restored trajectory of the moving object.

[0081] After obtaining the feasible roads of each acquisition device, the trajectory restoration device needs to connect the feasible roads to obtain the restored trajectory of the moving object. Figure 5, the dotted part in the figure is the road in the navigation map, the thinner solid line is the target movement trajectory, and the thicker solid line is the restored trajectory. When the target object moves from acquisition device A to B, the feasible roads determined by acquisition device A are Road A and Road B, and the feasible roads determined by acquisition device B are Road C. The target object can move from Road A to Road C at the intersection between Road A and Road C, or from Road B to Road C at the intersection between Road B and Road C. One of the methods can be selected to connect the feasible roads between acquisition device A and acquisition device B. The feasible roads of adjacent acquisition devices are connected in turn in the above manner to obtain the restored trajectory of the moving object. Exemplarily, the feasible paths between each acquisition device can be searched by the A-star algorithm. In some embodiments, the shortest path among the feasible paths between adjacent acquisition devices can also be selected to obtain the restored trajectory of the moving object, such as Figure 5 Indicated by the thicker solid line.

[0082] It can be seen that the trajectory restoration method of the embodiment of the present application combines the target movement trajectory with the navigation map to obtain the restored trajectory of the moving object, which can obtain a richer and more feasible moving object path.

[0083] Based on the above embodiments, the embodiments of the present application adopt Figure 6 The flowchart details how to predict the target's moving trajectory. Figure 6 , Figure 6 FIG. 1 is a flow chart of an exemplary embodiment of a trajectory prediction method shown in the present application. Specifically, the following steps are included:

[0084] S610: Acquire the target moving trajectory of the moving object.

[0085] The target moving trajectory is obtained by the above-mentioned trajectory completion method, and the main acquisition method is based on acquiring the captured images captured by multiple acquisition devices set in the target area; constructing the original moving trajectory of each moving object based on the captured images; in response to the missing trajectory of the original moving trajectory, the original moving trajectory is completed using the preset trajectory completion rules to obtain the target moving trajectory of the moving object, and the preset trajectory completion rules include a variety of trajectory completion methods arranged in the order of completion. After obtaining the completed trajectory, the future trajectory of the target moving trajectory can be predicted to obtain the predicted trajectory of the moving object.

[0086] The trajectory prediction device obtains the target moving trajectory of the moving object completed by the trajectory completion method.

[0087] S620: encoding each trajectory point in the target moving trajectory according to the acquisition time sequence of each trajectory point to obtain each encoded trajectory point.

[0088] When determining the trajectory point of a mobile object, it is necessary not only to determine the position of the trajectory point, but also to determine the collection time of the mobile object at the trajectory point, that is, the collection time when the collection device corresponding to the trajectory point collects the mobile object. The collection time order can be to sort the trajectory points from early to late according to the collection time, or to sort the trajectory points from late to early according to the collection time. For example, the earliest trajectory point can be sorted as 1, the second trajectory point can be sorted as 2, and so on. Afterwards, the geographical location, collection time order, etc. of each trajectory point of the target mobile trajectory can be encoded in the order of collection time to obtain each encoded trajectory point. Exemplarily, the encoding method can be one-hot encoding. The semantic encoding of each trajectory point is obtained by one-hot encoding.

[0089] S630: constructing a trajectory feature matrix according to the correlation between the trajectory points after the encoding process.

[0090] The association relationship can be the time relationship or spatial relationship of each encoded trajectory point in the target moving trajectory. The trajectory points obtained in step S620 after each encoding process are combined into a trajectory set, and the trajectory set is segmented using a time window. The semantic features of each trajectory point are constructed through the word embedding idea, and the association relationship between each encoded trajectory point is extracted, and a trajectory feature matrix is ​​constructed through the association relationship. Among them, the word embedding idea includes context trajectory combined with the field of natural language processing, such as Word2Vec, probabilistic language model and other algorithms.

[0091] The trajectory feature matrix includes the correlation between each trajectory point, such as the temporal relationship or spatial relationship between two trajectory points. Through the temporal relationship or spatial relationship, we can know whether they are connected in the target moving trajectory, which is 1 if connected and 0 if not connected.

[0092] The trajectory prediction device obtains the semantic features of each trajectory point after encoding, extracts the correlation between each trajectory point through the semantic features, and obtains the trajectory feature matrix through the correlation between each trajectory point.

[0093] S640: Inputting the trajectory feature matrix into the trained trajectory prediction model to obtain a predicted trajectory of the moving object output by the trained trajectory prediction model.

[0094] The trajectory prediction model is a trajectory prediction model trained by training samples, and the relationship between each trajectory point in the trajectory feature matrix is ​​learned through the trained trajectory prediction model. For example, a neural network similar to DBN can be used to learn the relationship between the hidden trajectory points in the trajectory feature matrix. This model can improve the generalization ability of trajectory features and express the spatiotemporal characteristics of each trajectory point in a deeper level. Based on the above trajectory feature matrix, a machine learning algorithm can be used to predict the target moving trajectory of the mobile object in real time. For example, regression models such as XGBoost, random forest, FM, and DeepFM.

[0095] Based on the above embodiments, the trajectory completion, restoration and prediction of the present application are described in detail:

[0096] See also Figure 7 and Figure 8 , Figure 7 It is a schematic diagram of the framework of the trajectory completion, restoration and prediction method shown in this application; Figure 8 : This is a schematic diagram of the framework of the trajectory completion, restoration and prediction method shown in this application. The framework includes an interactive page 710, a trajectory completion module 720, a heterogeneous verification module 730, a trajectory restoration module 740 and a trajectory prediction module 750. The trajectory restoration architecture based on trajectory completion, prediction as the core, and heterogeneity as verification provides multiple functions through the combination of multiple modules; it combines user interaction, structured data, view big data and deep learning together, with complete functions and simple interactions.

[0097] First, data preparation is performed to obtain images collected by multiple collection devices set in the target area. Interactive page 710 includes license plate number search, trajectory date selection, original movement trajectory, trajectory restoration, trajectory prediction, and vehicle daily trajectory restoration analysis buttons to meet special vehicle management needs, and also has image and video playback functions. Among them, trajectory prediction contains trajectory restoration analysis to improve trajectory density and can be used as real-time analysis; trajectory restoration can analyze each vehicle for any date; vehicle daily trajectory restoration analysis is the background real-time calculation of the target vehicle's restoration path and predicted trajectory, which does not rely on front-end click triggering.

[0098] The trajectory completion module 720 is the foundation of the entire framework, and aims to enrich the trajectory data by mining the potential passing points of the target vehicle by obtaining the mutual behavior rules between the vehicle and the area. The heterogeneous verification module 730 is used to verify the added trajectory points of the trajectory completion module 720. The trajectory completion methods include spatiotemporal analysis waste film recall, multi-algorithm crossover and similarity completion. The spatiotemporal analysis waste film recall includes high-frequency point waste film recall and high-frequency time period waste film recall; multi-algorithm crossover includes individual transition probability, global transition probability, trajectory anomaly detection and different moving object inspection based on fused data; similarity completion includes strong adjoint algorithm and trajectory similarity based on edit distance. Specifically, for example, the original moving trajectory is first completed by recalling waste films through spatiotemporal analysis to obtain a trajectory after the first completion processing; the heterogeneous verification module 730 obtains the trajectory points added in the trajectory after the first completion processing, verifies the added trajectory points, and returns the verification result to the trajectory completion module 720, and the trajectory completion module 720 removes the trajectory points that fail the verification to obtain the trajectory after the first processing; then the trajectory after the first processing is second completed by a probability-based completion algorithm to obtain a trajectory after the second completion processing; the heterogeneous verification module 730 obtains the trajectory points added in the trajectory after the second completion processing, verifies the added trajectory points, and returns the verification result to the trajectory completion module 720, and the trajectory completion module 720 removes the trajectory points that fail the verification to obtain the trajectory after the second processing; then the trajectory after the first processing is second completed by a probability-based completion algorithm The completion algorithm of the trajectory similarity performs a third completion process on the trajectory after the second processing to obtain the trajectory after the third completion process; the heterogeneous verification module 730 obtains the trajectory points added in the trajectory after the third completion process, verifies the added trajectory points, and returns the verification result to the trajectory completion module 720. The trajectory completion module 720 removes the trajectory points that fail the verification to obtain the trajectory after the third processing; the trajectory after the third processing is then completed for the fourth time through different moving object inspections based on the fused data to obtain the trajectory after the fourth completion process; the heterogeneous verification module 730 obtains the trajectory points added in the trajectory after the fourth completion process, verifies the added trajectory points, and returns the verification result to the trajectory completion module 720. The trajectory completion module 720 removes the trajectory points that fail the verification to obtain the target moving trajectory of the moving object.

[0099] The trajectory restoration module 740 obtains the target movement trajectory obtained by the trajectory completion module 720 and the heterogeneous verification module 730, combines the trajectory points of the target movement trajectory with the roads in the navigation map, and restores the target movement trajectory to the navigation map to form a complete path. Specifically, the target movement trajectory is subjected to feature extraction based on the word embedding concept to obtain the trajectory feature; the unsupervised deep learning trajectory feature is used to obtain the correlation between the trajectory points in the trajectory feature; the deep learning time series prediction model is used to input the unsupervised deep learning trajectory features into the deep learning time series prediction model for real-time prediction; and the predicted trajectory is then subjected to weight clustering optimization to obtain the optimized predicted trajectory.

[0100] The trajectory prediction module 750 constructs a trajectory feature matrix according to the correlation between each trajectory point in the target movement trajectory, inputs the trajectory feature matrix into the trained trajectory prediction model, and performs real-time prediction on the target movement trajectory to obtain a predicted trajectory.

[0101] In other embodiments, after the target movement trajectory is obtained through the trajectory completion module 720 and the heterogeneous verification module 730, a video verification function may be added to improve the accuracy of the trajectory completion.

[0102] In other embodiments, after the trajectory prediction module 750 performs trajectory prediction based on the target moving trajectory obtained by the trajectory completion module 720, the heterogeneous verification module 730 can be combined to perform verification processing on the added trajectory points in the predicted trajectory to obtain a verification result, and the predicted trajectory is updated according to the verification result.

[0103] See also Fig. 9 , Fig. 9 It is a structural schematic diagram of an exemplary embodiment of a trajectory completion device shown in the present application. The trajectory completion device 900 includes an acquisition module 910, a construction module 920 and a completion module 930; the acquisition module 910 is used to acquire images acquired by multiple acquisition devices set in the target area; the construction module 920 is used to construct the original movement trajectory of each moving object based on the acquired image; the completion module 930 is used to respond to the original movement trajectory There is a missing trajectory, and use a preset trajectory completion rule to complete the original movement trajectory to obtain the target movement trajectory of the mobile object. The preset trajectory completion rule includes a variety of trajectory completion methods arranged in order of completion.

[0104] In the above scheme, the trajectory completion method of the embodiment of the present application obtains the captured images captured by multiple capture devices set in the target area; constructs the original movement trajectory of each moving object based on the captured images; in response to the original movement trajectory having missing trajectories, the original movement trajectory is completed using a preset trajectory completion rule to obtain the target movement trajectory of the moving object, and the preset trajectory completion rule includes a plurality of trajectory completion methods arranged in a sequence of completion. Thus, the trajectory of the original movement trajectory with missing trajectories is completed using a plurality of trajectory completion methods in a sequence of completion, and the completion possibility is deeply explored from different data angles, thereby improving the completion accuracy.

[0105] The functions of each module can be found in the trajectory completion method embodiment, which will not be described in detail here.

[0106] In order to implement the trajectory completion method and / or trajectory restoration method and / or trajectory prediction method of the above-mentioned embodiment, the present application proposes another electronic device, which is specifically referred to as Fig.10 , Fig.10 It is a structural schematic diagram of an embodiment of an electronic device provided by the present application.

[0107] The electronic device 1000 includes a memory 1001 and a processor 1002 , wherein the memory 1001 and the processor 1002 are coupled.

[0108] The memory 1001 is used to store program data, and the processor 1002 is used to execute the program data to implement the trajectory completion method and / or trajectory restoration method and / or trajectory prediction method of the above-mentioned embodiments.

[0109] In this embodiment, the processor 1002 may also be referred to as a CPU (Central Processing Unit). The processor 1002 may be an integrated circuit chip having signal processing capabilities. The processor 1002 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. A general-purpose processor may be a microprocessor or the processor 1002 may also be any conventional processor, etc.

[0110] The present application also provides a computer-readable storage medium 1100, such as Fig.11 As shown, the computer-readable storage medium 1100 is used to store program data 1101. When the program data 1101 is executed by the processor, it is used to implement the trajectory completion method and / or trajectory restoration method and / or trajectory prediction method in the method embodiment of the present application.

[0111] The methods involved in the embodiments of the trajectory completion method and / or trajectory restoration method and / or trajectory prediction method of the present application, when implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a device, such as a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0112] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A trajectory completion method, characterized in that: The trajectory completion method comprises: Acquire captured images captured by multiple capture devices arranged in the target area; Constructing the original moving trajectory of each moving object based on the collected image; In response to the fact that the original moving trajectory is missing, a preset trajectory completion rule is used to complete the original moving trajectory to obtain a target moving trajectory of the moving object, wherein the preset trajectory completion rule includes performing serial trajectory completion on the original moving trajectory in a plurality of trajectory completion methods arranged in a completion order, wherein the plurality of trajectory completion methods include at least two of the following: spatiotemporal analysis waste film recall, probability-based individual and global completion algorithms, different moving object inspection based on fused data, license plate anomaly detection completion algorithm, trajectory completion based on an adjoint algorithm, and completion algorithm based on trajectory similarity; The preset trajectory completion rule includes a first trajectory completion mode and a second trajectory completion mode arranged in a completion order. The step of completing the original moving trajectory using the preset trajectory completion rule to obtain the target moving trajectory of the moving object includes: Performing a first processing on the original movement trajectory based on the first trajectory completion method to obtain a processed movement trajectory; The processed moving trajectory is processed for a second time based on the second trajectory completion method to obtain a target moving trajectory of the moving object, wherein the first trajectory completion method and the second trajectory completion method complete the trajectory by analyzing the acquisition image acquired by the acquisition device.

2. The trajectory completion method according to claim 1, characterized in that: The step of performing a first processing on the original movement trajectory based on the first trajectory completion method to obtain a processed movement trajectory includes: Completing the original movement trajectory based on the first trajectory completion method to obtain a completed movement trajectory; Performing verification processing on the added trajectory points in the completed moving trajectory to obtain a verification result; In response to the verification result indicating that there is no trajectory point with completion error in the added trajectory points, taking the completed movement trajectory as the processed movement trajectory; In response to the verification result indicating that there are completed incorrect trajectory points in the added trajectory points, the completed incorrect trajectory points are removed from the completed movement trajectory, and the movement trajectory after removing the incorrect trajectory points is used as the processed movement trajectory.

3. The trajectory completion method according to claim 2, characterized in that: The step of performing verification processing on the added trajectory points in the completed moving trajectory to obtain a verification result includes: Taking the moving object corresponding to the completed moving trajectory as the target object; Obtaining object features of a moving object in a video frame collected by a collection device within a collection time period corresponding to each added trajectory point, wherein the collection time period represents a collection time for the collection device to collect the target object; Calculating a first feature similarity between an object feature of the moving object and a target object feature of the target object; In response to the first feature similarity being greater than or equal to a first preset feature similarity threshold, characterizing the added trajectory point as a correct completed trajectory point; In response to the first feature similarity being less than the first preset feature similarity threshold, the added trajectory point is characterized as a completed erroneous trajectory point.

4. The trajectory completion method according to claim 2, characterized in that: The step of performing verification processing on the added trajectory points in the completed moving trajectory to obtain a verification result includes: Obtaining object features of a moving object in a video frame captured by a capture device within a capture time period corresponding to each added trajectory point, wherein the capture time period represents a capture time for the capture device to capture a target object, and the target object is a moving object corresponding to the completed moving trajectory; Determining a similarity scaling parameter of each moving object according to a spatiotemporal distance between each moving object and other trajectory points in the video frame, wherein the other trajectory points represent trajectory points in the original moving trajectory other than the added trajectory points; Determine the product of the similarity scaling parameter of each moving object and the corresponding first feature similarity as the second feature similarity of the corresponding moving object, wherein the first feature similarity represents the feature similarity between the object feature of each moving object and the target object feature of the target object; In response to the second feature similarity being greater than or equal to a first preset feature similarity threshold, characterizing the added trajectory point as a correct completed trajectory point; In response to the second feature similarity being less than the first preset feature similarity threshold, the added trajectory point is characterized as a completed erroneous trajectory point.

5. A trajectory restoration method, characterized in that: The trajectory restoration method comprises: Acquire a target moving trajectory of a mobile object, wherein the target moving trajectory is based on acquisition images acquired by a plurality of acquisition devices arranged in a target area; construct an original moving trajectory of each mobile object based on the acquisition images; in response to the presence of missing trajectories in the original moving trajectory, use a preset trajectory completion rule to complete the original moving trajectory to obtain the target moving trajectory of the mobile object, wherein the preset trajectory completion rule includes performing serial trajectory completion on the original moving trajectory using a plurality of trajectory completion methods arranged in a completion order, wherein the plurality of trajectory completion methods include at least two of the following: spatiotemporal analysis waste film recall, probability-based individual and global completion algorithms, different moving object inspections based on fused data, license plate anomaly detection completion algorithms, Trajectory completion based on adjoint algorithm and completion algorithm based on trajectory similarity; the preset trajectory completion rule includes a first trajectory completion method and a second trajectory completion method arranged in order of completion, the step of completing the original moving trajectory using the preset trajectory completion rule to obtain the target moving trajectory of the moving object includes: performing a first processing on the original moving trajectory based on the first trajectory completion method to obtain a processed moving trajectory; performing a second processing on the processed moving trajectory based on the second trajectory completion method to obtain the target moving trajectory of the moving object, wherein the first trajectory completion method and the second trajectory completion method perform trajectory completion by analyzing the captured images captured by the capture device; Determine feasible roads between adjacent collection devices from the navigation map according to actual installation positions of collection devices corresponding to each track point in the target moving track; The target moving track is restored to the navigation map according to the feasible road to obtain the restored track of the moving object.

6. The trajectory restoration method according to claim 5, characterized in that: The step of determining feasible roads between adjacent collection devices from a navigation map according to actual installation positions of collection devices corresponding to each track point in the target moving track comprises: In response to the actual installation location of the collection device being located on a road, determining the road corresponding to the collection device as a feasible road; In response to the actual installation location of the collection device not being located on a road, determining a road adjacent to the collection device as a feasible road according to an end point of the target moving trajectory; The feasible roads of the adjacent collection devices are connected at intersections between the feasible roads of the adjacent collection devices to obtain feasible roads between the adjacent collection devices.

7. A trajectory prediction method, characterized in that: The trajectory prediction method comprises: Acquire a target moving trajectory of a mobile object, wherein the target moving trajectory is based on acquisition images acquired by a plurality of acquisition devices arranged in a target area; construct an original moving trajectory of each mobile object based on the acquisition images; in response to the presence of missing trajectories in the original moving trajectory, use a preset trajectory completion rule to complete the original moving trajectory to obtain the target moving trajectory of the mobile object, wherein the preset trajectory completion rule includes performing serial trajectory completion on the original moving trajectory using a plurality of trajectory completion methods arranged in a completion order, wherein the plurality of trajectory completion methods include at least two of the following: spatiotemporal analysis waste film recall, probability-based individual and global completion algorithms, different moving object inspections based on fused data, license plate anomaly detection completion algorithms, Trajectory completion based on adjoint algorithm and completion algorithm based on trajectory similarity; the preset trajectory completion rule includes a first trajectory completion method and a second trajectory completion method arranged in order of completion, the step of completing the original moving trajectory using the preset trajectory completion rule to obtain the target moving trajectory of the moving object includes: performing a first processing on the original moving trajectory based on the first trajectory completion method to obtain a processed moving trajectory; performing a second processing on the processed moving trajectory based on the second trajectory completion method to obtain the target moving trajectory of the moving object, wherein the first trajectory completion method and the second trajectory completion method perform trajectory completion by analyzing the captured images captured by the capture device; Encoding each trajectory point according to the acquisition time sequence of each trajectory point in the target moving trajectory to obtain each encoded trajectory point; Construct a trajectory feature matrix according to the correlation between the trajectory points after each encoding process; The trajectory feature matrix is ​​input into a trained trajectory prediction model to obtain a predicted trajectory of the moving object output by the trained trajectory prediction model.

8. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores program instructions, and the processor retrieves the program instructions from the memory to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: include: Program data is stored, and when the program data is executed by a processor, it is used to implement the method according to any one of claims 1 to 7.

Images