Target tracking method, storage medium and autonomous vehicle
By obtaining the characteristic information of multi-frame environment data in unmanned driving, timing correlation and trajectory prediction are carried out, the target missed detection and misdetection problems caused by bicycle occlusion are solved, and more accurate and reliable target tracking is achieved, reducing the dependence on manual rules is achieved, and suitable for low-speed multi-occlusion scenarios.
Patent Information
- Application Number
- CN202510569970.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-08-12
AI Technical Summary
In the field of unmanned driving, bicycles or obstacles blocking lead to missed or missed targets. The existing multi-target tracking technology relies too much on manual rules, making it difficult to adapt to new goals and new scenarios, increase workload and is prone to missed detection.
By obtaining the characteristic information of multi-frame environment data, the initial tracking trajectory is constructed in timing correlation, the interrupted part is interpolated, overlapping trajectories are merged, and the trajectory filtering conditions are used to trigger the trajectory prediction mechanism to generate a continuous second tracking trajectory to reduce dependence on manual design.
It improves the accuracy and reliability of target tracking, reduces the dependence of traditional rules and methods, solves the trajectory splitting problem caused by complex occlusion, improves detection accuracy and recall, and has high versatility and practicality.
Smart Images
Figure CN120472367A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of unmanned driving, and specifically to a target tracking method, a storage medium, and an autonomous driving vehicle. Background Art
[0002] In the automatic labeling link in the field of unmanned driving, the vehicle or the barrier may block other targets to be detected, resulting in missed detection or false detection.
[0003] Currently, multi-target tracking technology primarily relies on traditional rules for target prediction to obtain the trajectory of the target to be detected. However, this approach has numerous problems: First, it relies too heavily on the design of artificial rules and fixed prior knowledge; second, when faced with new targets and scenarios, new rules often need to be redeveloped and adapted, which not only increases the workload but also easily leads to false detections and missed detections. Summary of the Invention
[0004] In view of this, embodiments of the present application provide a target tracking method, a storage medium, and an autonomous driving vehicle.
[0005] In a first aspect, an embodiment of the present application provides a target tracking method, including: obtaining feature information of a target to be detected contained in multiple frames of environmental data; based on the feature information of the target to be detected contained in the multiple frames of environmental data, performing time-series association on the same target to be detected in different frames of environmental data to obtain a first tracking trajectory of the same target to be detected; if the first tracking trajectory of the same target to be detected meets a trajectory screening condition, performing trajectory prediction on the same target to be detected to obtain a second tracking trajectory of the same target to be detected, wherein the trajectory screening condition includes at least one of the following items: the first tracking trajectory cannot cover the time range of all multiple frames of environmental data; there are at least two first tracking trajectories of the same target to be detected.
[0006] In combination with the first aspect, in certain implementations of the first aspect, based on the feature information of the target to be detected contained in multiple frames of environmental data, the same target to be detected in different frames of environmental data is temporally associated to obtain a first tracking trajectory of the same target to be detected, including: based on the feature information of the target to be detected contained in the multiple frames of environmental data, an initial tracking trajectory corresponding to the same target to be detected and containing a timing identifier is constructed; if there is a timing interruption in the initial tracking trajectory, the initial tracking trajectory is interpolated and completed to obtain the first tracking trajectory of the same target to be detected.
[0007] In combination with the first aspect, in certain implementations of the first aspect, the feature information of the target to be detected includes an instance identifier and a detection frame of the target to be detected. Based on the feature information of the target to be detected contained in multiple frames of environmental data, an initial tracking trajectory corresponding to the same target to be detected and including a timing identifier is constructed, including: based on the instance identifier and detection frame of the target to be detected in each frame of environmental data, a trace formed by the detection frame with the same instance identifier is used as the initial tracking trajectory of the same target to be detected; time-sorting the multiple frames of environmental data to obtain a timing sequence number for each frame of environmental data; and recording the timing sequence number of the environmental data where each detection frame on the initial tracking trajectory is located to obtain an initial tracking trajectory corresponding to the same target to be detected and including a timing identifier.
[0008] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: if the same target to be detected corresponds to multiple initial tracking trajectories, and there is time overlap between the multiple initial tracking trajectories, then the multiple initial tracking trajectories of the same target to be detected are merged to obtain only one initial tracking trajectory for the same target to be detected.
[0009] In combination with the first aspect, in certain implementations of the first aspect, trajectory prediction is performed on the same target to be detected to obtain a second tracking trajectory of the same target to be detected, including: predicting the trajectory position of the same target to be detected at a timing interruption based on the first tracking trajectory of the same target to be detected; matching the trajectory position of the same target to be detected at the timing interruption with the trajectory positions detected at other timings based on feature information of the target to be detected; and fusing the successfully matched trajectory position with the first tracking trajectory to generate a second tracking trajectory of the same target to be detected.
[0010] In conjunction with the first aspect, in certain implementations of the first aspect, the feature information of the target to be detected includes a detection frame of the target to be detected. Before obtaining the feature information of the target to be detected contained in multiple frames of environmental data, the method further includes: obtaining initial feature information of the target to be detected contained in the multiple frames of environmental data; filtering the initial feature information of the target to be detected by at least one of the following methods to obtain the feature information of the target to be detected: spatially filtering the detection frame of the target to be detected, retaining the detection frame within a predefined area; performing confidence filtering on the detection frame of the target to be detected based on a preset confidence threshold; and performing overlapping area filtering on the detection frame of the target to be detected based on a preset overlap rate threshold.
[0011] In combination with the first aspect, in certain implementations of the first aspect, the predefined area includes an area within a target distance range centered on the vehicle; and / or the confidence threshold is set based on the category of the target to be detected and / or the distance between the detection frame of the target to be detected and the vehicle; and / or the overlap rate threshold is set based on the category of the target to be detected.
[0012] In conjunction with the first aspect, in certain implementations of the first aspect, the multi-frame environmental data includes multi-frame image data and multi-frame point cloud data. Obtaining initial feature information of a target to be detected contained in the multi-frame environmental data includes: inputting the multi-frame image data and the multi-frame point cloud data into a multimodal sparse query network to obtain the initial feature information of the target to be detected, wherein the multimodal sparse query network is obtained through supervised training using instance identification of the multimodal data.
[0013] In the second aspect, an embodiment of the present application provides a target tracking device, including: an acquisition module for acquiring feature information of a target to be detected contained in multiple frames of environmental data; a timing association module for performing timing association on the same target to be detected in different frames of environmental data based on the feature information of the target to be detected contained in the multiple frames of environmental data, to obtain a first tracking trajectory of the same target to be detected; a trajectory detection module for performing trajectory prediction on the same target to be detected if the first tracking trajectory of the same target to be detected meets a trajectory screening condition, to obtain a second tracking trajectory of the same target to be detected, wherein the trajectory screening condition includes at least one of the following items: the first tracking trajectory cannot cover the timing range of all multiple frames of environmental data; there are at least two first tracking trajectories of the same target to be detected.
[0014] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program for executing the target tracking method described in the first aspect.
[0015] In a fourth aspect, an embodiment of the present application provides an autonomous driving vehicle, which includes: a processor; a memory for storing processor-executable instructions; and the processor is used to execute the target tracking method described in the first aspect.
[0016] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes instructions that, when executed on an autonomous driving vehicle, enable the autonomous driving vehicle to implement the target tracking method described in the first aspect.
[0017] In the present application, in the time series association stage, the target to be detected in the multi-frame environmental data is accurately matched through feature information to generate a first tracking trajectory, thereby enhancing the accuracy and reliability of target tracking. When the target to be detected is lost for a long time, resulting in the first tracking trajectory being unable to cover the entire time series range, or when the same target to be detected is mistakenly detected as multiple split trajectories, the trajectory prediction mechanism is triggered based on the preset trajectory screening conditions, thereby restoring or merging the split trajectories to form a continuous second tracking trajectory. In addition, while improving the detection accuracy and recall rate, this method reduces the dependence of traditional rule methods on manual design through a hierarchical processing mechanism, avoids manual parameter adjustments for different categories or environments of the target to be detected, and collaboratively solves the problem of trajectory splitting caused by complex occlusions. This solution can be extended to other autonomous driving systems for low-speed and multi-occlusion scenarios, and has high versatility and practicality. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.
[0019] Figure 1 The figure is a flow chart of a target tracking method provided in one embodiment of the present application.
[0020] Figure 2 FIG2 is a flow chart of determining a first tracking trajectory of the same target to be detected provided by an embodiment of the present application.
[0021] Figure 3 FIG2 is a flow chart of determining a second tracking trajectory of the same target to be detected provided by an embodiment of the present application.
[0022] Figure 4 Shown is a flow chart of a target tracking method provided in another embodiment of the present application.
[0023] Figure 5 Shown is a schematic structural diagram of a target tracking device provided in one embodiment of the present application.
[0024] Figure 6 Shown is a schematic structural diagram of an autonomous driving vehicle provided in one embodiment of the present application. DETAILED DESCRIPTION
[0025] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0026] Figure 1 FIG. 1 is a flow chart of a target tracking method provided by an embodiment of the present application. Figure 1 As shown, the method includes the following steps.
[0027] Step S110: Acquire feature information of the target to be detected contained in multiple frames of environmental data.
[0028] Multi-frame environmental data refers to a series of environmental information acquired at different time points during target tracking. For example, in autonomous driving scenarios, vehicles are equipped with multiple sensors (such as lidar, cameras, and millimeter-wave radars), which continuously acquire information about the surrounding environment. This series of data acquired at different time points constitutes multi-frame environmental data.
[0029] Detectable targets are objects that need to be tracked within these environmental data. Optionally, these targets include entities with certain characteristics, such as pedestrians, vehicles, traffic signs, and traffic lights. Feature information is used to describe the various attributes and characteristics of the detected targets, such as their shape, color, size, outline, texture, identification, and location information within multiple frames of environmental data. This helps identify and distinguish different detected targets, as well as the state changes of the same target across different frames.
[0030] In one implementation, for each target to be detected, multiple frames of environmental data are analyzed and processed (such as image recognition, data fusion, etc.) to extract the feature information corresponding to the target to be detected in each frame of environmental data, so as to facilitate the subsequent association of data of the same target to be detected at different time points and construct its tracking trajectory.
[0031] Step S120 , based on the feature information of the target to be detected contained in the multiple frames of environmental data, time-series correlation is performed on the same target to be detected in different frames of environmental data to obtain a first tracking trajectory of the same target to be detected.
[0032] The same target refers to instances of the same physical entity at different time points across multiple frames of environmental data. Temporal correlation involves connecting and matching instances of the same target across different frames of environmental data in chronological order based on the target's feature information, thereby establishing a continuous relationship between the targets over time.
[0033] The first tracking trajectory reflects the path of the same target being detected over a period of time. It is understandable that the first tracking trajectory may still have some imperfections, such as interruptions in the trajectory and lack of accuracy, which require further processing and optimization.
[0034] In one implementation, the similarity between the targets to be detected in different frames of environmental data is calculated based on the feature information of the targets to be detected, such as appearance features and motion features. Furthermore, a similarity threshold is determined. When the similarity between the feature information of the targets to be detected in two frames of environmental data exceeds the similarity threshold, the targets to be detected in the two frames of environmental data are considered to be the same targets to be detected, thereby performing temporal association. For example, for vehicle tracking in an autonomous driving scenario, it is possible to determine whether the vehicles in different frames of environmental data are the same vehicle by comparing the vehicle's appearance features such as color and model, as well as the vehicle's motion features such as speed and direction, thereby obtaining the vehicle's first tracking trajectory.
[0035] Step S130 : If the first tracking trajectory of the same target to be detected meets the trajectory screening condition, trajectory prediction is performed on the same target to be detected to obtain a second tracking trajectory of the same target to be detected.
[0036] The trajectory screening condition is used to determine whether the first tracking trajectory needs further processing, including at least one of the following: the first tracking trajectory cannot cover the time series range of all multi-frame environmental data; there are at least two first tracking trajectories for the same target to be detected.
[0037] Specifically, when the first tracking trajectory cannot cover the temporal range of all multi-frame environmental data, it means that there is a temporal discontinuity or omission in the trajectory. This may be because the target to be detected was not detected at certain moments (such as being blocked or the sensor was lost), resulting in an incomplete trajectory that cannot fully reflect the movement of the target to be detected in the entire time series. Secondly, if there are at least two first tracking trajectories for the same target to be detected, this indicates that there has been a problem of trajectory splitting or repeated detection during the tracking process. This may be due to the complex movement of the target to be detected, changes in appearance, errors in the detection algorithm, etc., which leads to the erroneous identification of the same target to be detected as multiple different trajectories.
[0038] These two screening conditions can help identify the first tracking trajectory that needs to be optimized or corrected, so as to generate a more accurate and complete second tracking trajectory through means such as trajectory prediction, thereby improving the accuracy and reliability of tracking the target to be detected, and ensuring that the solution in this application can effectively handle target tracking tasks in complex scenarios.
[0039] In one implementation, based on the historical trajectory information and motion model of the target to be detected, the trajectory of the same target to be detected is predicted. More specifically, the motion state of the target to be detected in the current frame environmental data can be predicted based on its motion state (position, speed, etc.) in the historical frame environmental data. Alternatively, with the help of a deep learning model, the feature information, historical trajectory information and motion model of the target to be detected are input, and a predicted second trajectory is output. It can be understood that the second tracking trajectory can overcome some of the defects of the first tracking trajectory, more completely cover the motion path of the target to be detected in multiple frames of environmental data, and there is only one second tracking trajectory for the same target to be detected, thereby improving the accuracy and consistency of tracking the target to be detected.
[0040] In this embodiment, in the time series association stage, the target to be detected in the multi-frame environmental data is accurately matched through feature information to generate a first tracking trajectory, thereby enhancing the accuracy and reliability of target tracking. When the target to be detected is lost for a long time, resulting in the first tracking trajectory being unable to cover the entire time series range, or when the same target to be detected is mistakenly detected as multiple split trajectories, the trajectory prediction mechanism is triggered based on the preset trajectory screening conditions, thereby restoring or merging the split trajectories to form a continuous second tracking trajectory. In addition, while improving the detection accuracy and recall rate, this method reduces the dependence of traditional rule methods on manual design through a hierarchical processing mechanism, avoids manual parameter adjustments for different categories or environments of the target to be detected, and collaboratively solves the problem of trajectory splitting caused by complex occlusions. This solution can be extended to other autonomous driving systems for low-speed and multi-occlusion scenarios, and has high versatility and practicality.
[0041] In some embodiments, in addition to Figure 1 In addition to the implementation methods for determining the initial feature information of the target to be detected listed in the embodiment, this embodiment also provides another method. Specifically, the multi-frame environmental data includes multiple frames of image data and multiple frames of point cloud data. The multiple frames of image data and multiple frames of point cloud data are input into a multimodal sparse query network to obtain the initial feature information of the target to be detected.
[0042] The multimodal sparse query network is obtained through supervised training of instance identification of multimodal data (including image data and point cloud data). Specifically, in the training phase of the multimodal sparse query network, the preprocessed multi-channel sample images and multi-channel sample point clouds are used as data sets. At the same time, the detection frames and instance identifications of the sample targets in the multi-channel sample images and multi-channel sample point clouds are used as labels and input into the multimodal sparse query network. It is trained using a large-scale data set to enable it to learn effective model weights. In the inference phase, multi-frame image data and multi-frame point cloud data are input into the trained multimodal sparse query network, and the detection frame containing the target to be detected and the corresponding instance identification can be output.
[0043] In this embodiment, training with instance identification as the supervisory signal can help the multimodal sparse query network learn the feature representation that distinguishes different target instances, enhance the recognition ability of the target to be detected, and accurately identify and track multiple targets to be detected even in complex scenarios, thereby improving tracking accuracy. Accordingly, the multimodal sparse query network can provide effective instance identification for target tracking, making tracking simple and efficient. In addition, querying in a sparse manner can efficiently extract key features related to the target to be detected from the multimodal data, avoid redundant information interference, improve feature extraction efficiency and pertinence, and enable the multimodal sparse query network to focus more on the key attributes of the target to be detected, thereby improving the accuracy and speed of identifying the target to be detected. Finally, in order to obtain a more accurate detection frame, multimodal data is used as the input of the multimodal sparse query network, avoiding the problem that the output detection frame of the single-modal input lacks texture or depth information, resulting in poor tracking accuracy, making target tracking simpler and more efficient.
[0044] Figure 2 The figure shows a flow chart of determining the first tracking trajectory of the same target to be detected provided by an embodiment of the present application. Figure 1 Based on the embodiment shown, Figure 2 The embodiment shown is described below in detail. Figure 2 The embodiment shown and Figure 1 The differences and similarities between the illustrated embodiments are not described in detail.
[0045] like Figure 2 As shown, in this embodiment, based on the feature information of the target to be detected contained in multiple frames of environmental data, the same target to be detected in different frames of environmental data is temporally associated to obtain the first tracking trajectory of the same target to be detected, including the following steps.
[0046] Step S210 : constructing an initial tracking trajectory corresponding to the same target to be detected and including a time sequence identifier based on the feature information of the target to be detected contained in the multiple frames of environmental data.
[0047] The timing identifier is used to indicate the order in which the target to be detected appears in different frames of environmental data. Optionally, the timing identifier includes a timestamp or frame number, which can help determine the movement direction and time sequence of the target to be detected in multiple frames of environmental data.
[0048] The initial tracking trajectory represents the original trajectory formed by preliminary matching of the same target to be detected across multiple frames of environmental data. This preliminary matching refers to roughly associating instances of the target to be detected across multiple frames of environmental data to form the original trajectory. It is understood that preliminary matching can quickly establish the association between the target to be detected across different frames of environmental data, but it may contain some error or uncertainty.
[0049] In one implementation, based on the characteristic information of the target to be detected, instances belonging to the same target to be detected in different frames of environmental data are identified. Optionally, during the identification process, a timing identifier is assigned to each target to be detected, which records the time sequence in which the target to be detected appears in the corresponding frame of environmental data. Then, the instances with the same characteristic information are sequentially associated in the order of the timing identifiers to form an initial tracking trajectory. For example, if a target to be detected is detected in the first frame of environmental data, the second frame of environmental data, and the third frame of environmental data, and the similarity of its characteristic information is greater than the similarity threshold, and the timing identifiers are incremented in sequence, then the instances in the three frames can be associated to construct the initial tracking trajectory of the target to be detected from the first frame of environmental data to the third frame of environmental data.
[0050] Step S220 : If there is a timing interruption in the initial tracking trajectory, the initial tracking trajectory is interpolated and completed to obtain a first tracking trajectory of the same target to be detected.
[0051] Timing interruption refers to the absence of a time series in the initial tracking trajectory. That is, in multi-frame environmental data, the detection box of the same target to be detected has a temporal discontinuity between some frames, resulting in the initial tracking trajectory being unable to fully reflect the movement of the target to be detected within that time period.
[0052] Interpolation completion involves constructing new data points between known data points to fill in missing data points, making the data sequence more complete and continuous. In object tracking, this involves predicting the possible location of the target during the time series interruption based on the known position and time information of the detection box, thereby completing the initial tracking trajectory.
[0053] In one implementation, based on the trajectory position of the same target to be detected in the initial tracking trajectory, before and / or after the timing interruption, the trajectory position of the same target to be detected at the timing interruption is completed by an interpolation algorithm. For example, assuming that there is a detection frame A at t1 and a detection frame C at t3 in the initial tracking trajectory, and the detection frame is missing at time t2 (located between t1 and t3). Then, the position of the detection frame B corresponding to time t2 can be calculated based on the detection frame A at time t1 before the timing interruption and the detection frame C at time t2 after the timing interruption.
[0054] It should be noted that the interpolation completion algorithm is mainly used to deal with short-term loss situations, because at this time there are relatively close and accurate known trajectory positions before and after the timing interruption for reference. Based on the movement trend and state of the target before and after the loss, the trajectory position at the timing interruption can be more accurately inferred. When the loss is long-term, the correlation between the known trajectory position and the actual movement of the target is greatly reduced, the reference value is weakened, and the interpolation algorithm is difficult to accurately predict the trajectory changes within a long time interval. In addition, long-term loss means that more unknown points need to be interpolated, and the error accumulation in the interpolation process is more significant, which may lead to a significant decrease in trajectory accuracy. Therefore, interpolation completion is more suitable for short-term loss scenarios.
[0055] Optionally, after interpolating and completing the initial tracking trajectory, the process also includes assigning corresponding time series identifiers to the interpolated targets to be detected. This ensures the integrity and traceability of the trajectory in the temporal dimension, enhances trajectory coherence, and provides a more precise time reference for subsequent data analysis and decision-making, helping to improve the accuracy and reliability of the entire target tracking process.
[0056] In this embodiment, the feature information of the target to be detected from multiple frames of environmental data is first constructed, including a timing identifier. Then, if any timing interruptions are found in the initial trajectory, interpolation is used to complete the trajectory, making it more complete and continuous, thereby more realistically reflecting the target's motion. Overall, the solution in this embodiment effectively improves the accuracy and reliability of target tracking and enhances its adaptability to complex scenarios.
[0057] In order to further improve the accuracy of constructing the initial tracking trajectory, another implementation method of constructing the initial tracking trajectory corresponding to the same target to be detected and containing a timing identifier based on the feature information of the target to be detected contained in the multi-frame environmental data is introduced below. Specifically, the feature information of the target to be detected includes the instance identifier and the detection frame of the target to be detected. Among them, the instance identifier is a specific label or number used to uniquely identify the target to be detected, which can distinguish different target individuals, so that it is possible to quickly determine which targets to be detected are the same in the multi-frame environmental data. The detection frame is a bounding box that surrounds the target to be detected in each frame of environmental data. Coordinate parameters (such as the upper left corner and lower right corner coordinates) are usually used to define the position range of the target to be detected in the frame of environmental data.
[0058] Furthermore, based on the instance identifier and detection frame of the target to be detected in each frame of environmental data, the trace formed by the detection frames with the same instance identifier is used as the initial tracking trajectory of the same target to be detected; multiple frames of environmental data are time-sorted to obtain a time sequence number for each frame of environmental data; the time sequence number of the environmental data where each detection frame is located on the initial tracking trajectory is recorded to obtain the initial tracking trajectory corresponding to the same target to be detected, including the time sequence identifier. The time sequence number is used to indicate the position of the frame environmental data in the entire time series, which helps to accurately reflect the movement sequence of the target to be detected when constructing the trajectory.
[0059] For example, in an autonomous driving scenario, the vehicle's sensors acquire three consecutive frames of environmental data, each corresponding to a time series identifier t1, t2, and t3. Furthermore, in the first frame (t1), vehicle X is detected with an instance identifier of "1" and a detection box coordinate of [100, 100, 200, 200] (indicates that in the image coordinate system, the upper left corner is (100, 100) and the lower right corner is (200, 200)). Vehicle Y is detected with an instance identifier of "2" and a detection box coordinate of [300, 150, 400, 250]. In the second frame (t2), vehicles X and Y are detected, their instance identifiers still "1" and "2," and their corresponding detection box coordinates are [150, 150, 250, 250] and [350, 200, 450, 300], respectively. In the third frame (t3), vehicle X (instance identified as “1”) and vehicle Y (instance identified as “2”) are detected, and the corresponding detection box coordinates are [200, 200, 300, 300] and [400, 250, 500, 350] respectively.
[0060] Furthermore, based on the aforementioned information, the traces formed by the detection boxes with the same instance identifier are used as the initial tracking trajectory of the same target to be detected. Here, the initial tracking trajectory of vehicle X consists of the detection boxes with instance identifier "1" in the first, second, and third frames; the initial tracking trajectory of vehicle Y consists of the detection boxes with instance identifier "2" in the first, second, and third frames.
[0061] Finally, record the time sequence number of the frame environment data where each detection box on the initial tracking trajectory is located. For example, the initial tracking trajectory of vehicle X: t1, detection box coordinates [100, 100, 200, 200]; t2, detection box coordinates [150, 150, 250, 250]; t3, detection box coordinates [200, 200, 300, 300]. The initial tracking trajectory of vehicle Y: t1, detection box coordinates [300, 150, 400, 250]; t2, detection box coordinates [350, 200, 450, 300]; t3, detection box coordinates [400, 250, 500, 350].
[0062] In this implementation, the uniqueness of instance identifiers enables accurate re-identification and re-association after the target to be detected temporarily disappears, thereby maintaining the continuity of target tracking and significantly improving the robustness of target tracking. This is particularly effective in addressing the challenges of false detection and missed detection in mining scenarios. Furthermore, traditional tracking methods typically require the design of a large number of complex tracking and matching rules for different scenarios and target categories. However, this solution, by introducing instance identifiers and detection boxes, can automatically perform target matching and tracking, achieving more versatile and efficient tracking results in different scenarios, greatly reducing the workload and complexity of manually designing rules.
[0063] In some embodiments, if the same target to be detected corresponds to multiple initial tracking trajectories and there is temporal overlap between the multiple initial tracking trajectories, the multiple initial tracking trajectories of the same target to be detected are merged to obtain only one initial tracking trajectory of the same target to be detected.
[0064] Temporal overlap means that two or more initial tracking trajectories have overlapping parts on the time axis, that is, they are recorded in some of the same time periods, which means that these trajectories actually correspond to different observation segments of the same target to be detected.
[0065] If time overlap is detected between multiple initial tracking tracks, these overlapping tracks need to be merged. During the merging process, the initial tracking tracks can be sorted chronologically, and their detection box information can be integrated into a single track. Optionally, the detection boxes within the overlapping time period need further processing, such as averaging and selecting the optimal detection box, to ensure the continuity and accuracy of the merged track within that time period and avoid duplicate recording.
[0066] For example, a vehicle sensor collects five consecutive frames of data, and the same target to be detected corresponds to two initial tracking trajectories: initial tracking trajectory 1 and initial tracking trajectory 2. Initial tracking trajectory 1 covers frames 1 to 3, while initial tracking trajectory 2 covers frames 3 to 5. Both instances have the same instance identifier of "1." Because the third frame overlaps in time, it needs to be merged. Since the detection frame of the third frame in initial tracking trajectory 2 is optimal, the detection frames of frames 1 and 2 of initial tracking trajectory 1 and frames 3, 4, and 5 of initial tracking trajectory 2 are retained after merging, forming a complete trajectory covering frames 1 to 5.
[0067] In this embodiment, by merging multiple initial tracking trajectories with temporal overlap into one, trajectory splitting caused by detection errors or occlusions can be eliminated, and an initial tracking trajectory that more completely and accurately reflects the motion of the same target to be detected can be obtained.
[0068] Figure 3 The figure shows a flow chart of determining a second tracking trajectory of the same target to be detected provided by an embodiment of the present application. Figure 1 Based on the embodiment shown, Figure 3 The embodiment shown is described below in detail. Figure 3 The embodiment shown and Figure 1 The differences and similarities between the illustrated embodiments are not described in detail.
[0069] like Figure 3 As shown, in this embodiment, performing trajectory prediction on the same target to be detected to obtain a second tracking trajectory of the same target to be detected includes the following steps.
[0070] Step S310 : predicting the trajectory position of the same target to be detected at the timing interruption based on the first tracking trajectory of the same target to be detected.
[0071] Optionally, in a case where the first tracking trajectory cannot cover the time sequence range of multiple frames of environmental data, the trajectory position of the target to be detected at the time sequence interruption is predicted based on the first tracking trajectory.
[0072] For example, based on the first tracking trajectory, the position, velocity, and acceleration of the target are determined and used as state variables. Then, based on the historical state vector and the target's motion model, the target's trajectory position at the time sequence interruption is predicted. When the target is re-detected, the newly observed state variables are used to correct the predicted target's trajectory position.
[0073] Step S320 , based on the feature information of the target to be detected, the trajectory position of the same target to be detected at the time sequence interruption is matched with the trajectory positions detected at other time sequences.
[0074] Optionally, the feature information includes a detection box and an instance identifier. By combining the position information of the detection box with the uniqueness of the instance identifier, the trajectory position of the same target to be detected at the time sequence interruption is matched with the trajectory positions detected at other time sequences.
[0075] Specifically, the detection box provides the position and size information of the target to be detected in the image or space, while the instance identifier assigns a unique identifier to each target to be detected, ensuring that different targets can be distinguished in multi-target scenarios. Calculate the matching cost (such as the Euclidean distance) between the position information of the predicted detection box (i.e., the trajectory position) and the newly observed detection box of the target to be detected, and find it.
[0076] Step S330 : Fusing the successfully matched trajectory position with the first tracking trajectory to generate a second tracking trajectory of the same target to be detected.
[0077] Optionally, the successfully matched trajectory position refers to a matching solution with the minimum cost. Specifically, once the trajectory position predicted at the timing interruption is successfully matched with the trajectory positions detected at other timings, these matched trajectory positions are integrated into the first tracking trajectory.
[0078] Optionally, during the fusion process, the positions of successfully matched tracks are inserted into the corresponding positions in the first tracking track to fill in the gaps in the timing sequence. To eliminate track discontinuities caused by the insertion of new positions, a smoothing algorithm can be used to smooth the entire track to ensure track consistency and fluidity. Finally, a more complete and accurate second tracking track is generated, thereby improving the accuracy and reliability of target tracking.
[0079] In this embodiment, the predicted trajectory position is matched with the trajectory positions detected at other time points using the characteristic information of the target to be detected. This ensures the correlation between the predicted trajectory position and the actual detected trajectory position, thereby verifying the accuracy of the prediction. The successfully matched trajectory position is then fused with the first tracking trajectory to generate a second tracking trajectory. This can reintegrate the trajectory that was originally broken due to long-term loss, solving the problem of trajectory fragmentation caused by the long-term loss of the target to be detected and ensuring the coherence of the second tracking trajectory.
[0080] Figure 4 FIG. 1 is a flow chart of a target tracking method provided by another embodiment of the present application. Figure 1 Based on the embodiment shown, Figure 4 The embodiment shown is described below in detail. Figure 4 The embodiment shown and Figure 1 The differences and similarities between the illustrated embodiments are not described in detail.
[0081] like Figure 4 As shown, in this embodiment, the feature information of the target to be detected includes the detection frame of the target to be detected. Before obtaining the feature information of the target to be detected contained in multiple frames of environmental data, the following steps are also included.
[0082] Step S410: obtaining initial feature information of the target to be detected contained in multiple frames of environmental data.
[0083] Optionally, the multiple frames of environmental data include multiple frames of image data and multiple frames of point cloud data.
[0084] In one implementation, multiple frames of image data are processed by a target detection algorithm to identify information such as the planar position of the target to be detected. At the same time, the depth information of the target to be detected is extracted from the multiple frames of point cloud data. Then, the initial detection frame of the target to be detected, that is, the initial feature information of the target to be detected, is determined based on the planar position and depth information.
[0085] Step S420 , performing at least one of the following filtering on the initial feature information of the target to be detected to obtain feature information of the target to be detected.
[0086] Optionally, the detection frame of the target to be detected is spatially filtered to retain the detection frame within a predefined area, where the predefined area includes an area within a target distance range centered on the vehicle.
[0087] Specifically, a predefined region centered on the ego vehicle is set, such as the target distance range. For example, the target distance range can be determined based on scenario requirements and mission objectives. Only detection frames within this predefined region are then retained. This effectively removes detection frames that are too far away or have a weak correlation with the ego vehicle, allowing subsequent processing to focus more on potential targets related to the ego vehicle, improving the relevance and efficiency of target detection and reducing interference from irrelevant information.
[0088] Optionally, the detection frame of the target to be detected is confidence filtered based on a preset confidence threshold. The confidence threshold is set based on the category of the target to be detected and / or the distance between the detection frame of the target to be detected and the vehicle.
[0089] Confidence is a quantitative assessment of the reliability of the detection results by the object detection algorithm. It's understood that the confidence threshold setting is not fixed and can be adjusted based on the type of target being detected. For example, a higher confidence threshold can be set for target categories prone to false detection. The distance between the detection bounding box of the target being detected and the ego vehicle can also be considered. Generally, the farther away from the ego vehicle, the greater the uncertainty of detection. Therefore, a relatively high confidence threshold can be set for the detection bounding box of distant targets.
[0090] Through this confidence filtering, detection frames with low confidence and poor reliability can be screened out, and more reliable target information can be retained, which helps to improve the accuracy of subsequent target tracking.
[0091] Optionally, based on a preset overlap rate threshold, the detection frame of the target to be detected is filtered for overlapping areas. The overlap rate threshold is set based on the category of the target to be detected.
[0092] The overlap rate refers to the ratio of the area of the overlapping region to the area of the detection frame. In addition, in this embodiment, the overlap rate threshold is also set based on the category of the target to be detected. Different categories of targets have different shapes, sizes and detection characteristics, and the tolerance for overlap is also different. For example, for targets such as pedestrians, since their shapes are relatively regular, the overlap rate threshold can be set relatively high; while for some irregularly shaped targets, a lower overlap rate threshold needs to be set. Furthermore, for detection frames whose overlap rate exceeds the overlap rate threshold, they are deleted to eliminate redundant repeated detection frames. Such overlapping area filtering can effectively solve the common problem of detection frame overlap in target detection, avoid a target being repeatedly detected by multiple detection frames, improve the clarity and accuracy of the detection results, and lay a good foundation for subsequent target tracking.
[0093] From a broader perspective, spatial filtering is the most direct screening method. It only retains detection frames within a predefined area, determined based on scenario requirements and mission objectives. This quickly eliminates detection frames that are clearly irrelevant to the current task or too far away, reducing the amount of data to be processed. Furthermore, spatially filtered detection frames are more focused on potential targets related to the ego vehicle, improving the targeted nature of target detection.
[0094] Confidence filtering is then performed. The confidence threshold at this time can be set based on the category of the target to be detected and the distance between the detection frame and the vehicle. This step reduces invalid detection frames while ensuring that the retained detection frames are more reliable, providing more accurate basic information for subsequent target tracking.
[0095] Finally, overlapping regions are filtered. At this point, the detection frames have been double-screened by spatial and confidence filtering, resulting in a relatively small number of them with higher confidence. By setting a reasonable overlap threshold, we can prevent a single target from being repeatedly detected by multiple detection frames, further improving the clarity and accuracy of the detection results.
[0096] In summary, by performing spatial filtering, confidence filtering, and overlapping area filtering in this order, the previous step provides the subsequent step with more targeted, fewer, and higher-quality detection frames, reducing the computational burden of subsequent steps and improving processing speed. Furthermore, this step-by-step filtering approach eliminates as many obviously invalid detection frames as possible in the early steps, while retaining those with potentially valid information. This ensures that the remaining detection frames are both accurate and efficient, thereby retaining more valid detection frames and filtering out more invalid ones.
[0097] Combined with the above Figures 1 to 4 , describes in detail the target tracking method embodiment of the present application, and the following is combined with Figure 5, the target tracking device embodiment of the present application is described in detail. It should be understood that the description of the target tracking method embodiment corresponds to the description of the target tracking device embodiment, so the parts not described in detail can be referred to the previous method embodiment.
[0098] Figure 5 The figure shows a schematic diagram of the structure of a target tracking device provided by an embodiment of the present application. Figure 5 As shown, the target tracking device 50 provided in the embodiment of the present application includes:
[0099] An acquisition module 510 is configured to acquire feature information of a target to be detected contained in multiple frames of environmental data;
[0100] A time series association module 520 is configured to perform time series association on the same target to be detected in different frames of environmental data based on feature information of the target to be detected contained in the multiple frames of environmental data, to obtain a first tracking trajectory of the same target to be detected;
[0101] The trajectory detection module 530 is configured to perform trajectory prediction on the same target to be detected to obtain a second tracking trajectory of the same target to be detected if the first tracking trajectory of the same target to be detected satisfies a trajectory screening condition, wherein the trajectory screening condition includes at least one of the following: the first tracking trajectory cannot cover the time series range of all multiple frames of environmental data; or there are at least two first tracking trajectories of the same target to be detected.
[0102] In one embodiment of the present application, the timing association module 520 is further used to construct an initial tracking trajectory corresponding to the same target to be detected and including a timing identifier based on the feature information of the target to be detected contained in multiple frames of environmental data; if there is a timing interruption in the initial tracking trajectory, the initial tracking trajectory is interpolated and completed to obtain a first tracking trajectory of the same target to be detected.
[0103] In one embodiment of the present application, the feature information of the target to be detected includes the instance identifier and detection frame of the target to be detected. The temporal association module 520 is further configured to, based on the instance identifier and detection frame of the target to be detected in each frame of environmental data, use the trace formed by the detection frames with the same instance identifier as the initial tracking trajectory of the same target to be detected; perform temporal sorting on multiple frames of environmental data to obtain a temporal sequence number for each frame of environmental data; and record the temporal sequence number of the environmental data where each detection frame on the initial tracking trajectory is located to obtain an initial tracking trajectory corresponding to the same target to be detected, including the temporal identifier.
[0104] In one embodiment of the present application, the timing association module 520 is further configured to, if the same target to be detected corresponds to multiple initial tracking trajectories and there is temporal overlap between the multiple initial tracking trajectories, merge the multiple initial tracking trajectories of the same target to be detected to obtain only one initial tracking trajectory of the same target to be detected.
[0105] In one embodiment of the present application, the trajectory detection module 530 is further used to predict the trajectory position of the same target to be detected at the timing interruption based on the first tracking trajectory of the same target to be detected; match the trajectory position of the same target to be detected at the timing interruption with the trajectory position detected at other timings based on the feature information of the target to be detected; and fuse the successfully matched trajectory position with the first tracking trajectory to generate a second tracking trajectory of the same target to be detected.
[0106] In one embodiment of the present application, the feature information of the target to be detected includes a detection frame of the target to be detected. The acquisition module 510 is further configured to obtain initial feature information of the target to be detected contained in multiple frames of environmental data; and to filter the initial feature information of the target to be detected by performing at least one of the following operations to obtain the feature information of the target to be detected: spatially filtering the detection frame of the target to be detected to retain the detection frame within a predefined area; performing confidence filtering on the detection frame of the target to be detected based on a preset confidence threshold; and performing overlapping area filtering on the detection frame of the target to be detected based on a preset overlap rate threshold.
[0107] In one embodiment of the present application, the acquisition module 510 is also used to predefine an area including an area within a target distance range centered on the vehicle; and / or, a confidence threshold is set based on the category of the target to be detected and / or the distance between the detection frame of the target to be detected and the vehicle; and / or, an overlap rate threshold is set based on the category of the target to be detected.
[0108] In one embodiment of the present application, the multi-frame environmental data includes multi-frame image data and multi-frame point cloud data. The acquisition module 510 is further configured to input the multi-frame image data and multi-frame point cloud data into a multimodal sparse query network to obtain initial feature information of the target to be detected, wherein the multimodal sparse query network is obtained through supervised training using instance identification of the multimodal data.
[0109] Below, reference Figure 6 To describe the autonomous driving vehicle according to an embodiment of the present application. Figure 6 Shown is a schematic structural diagram of an autonomous driving vehicle provided by an exemplary embodiment of the present application.
[0110] like Figure 6 As shown, the autonomous vehicle 60 includes one or more processors 601 and memory 602.
[0111] The processor 601 may be a central processing unit (CPU) or other form of processing unit having data processing capabilities and / or instruction execution capabilities, and may control other components in the autonomous driving vehicle 60 to perform desired functions.
[0112] The memory 602 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 601 may execute the program instructions to implement the target tracking method of each embodiment of the present application described above and / or other desired functions. Various contents such as feature information of the target to be detected, a first tracking trajectory, a second tracking trajectory, etc. may also be stored in the computer-readable storage medium.
[0113] In one example, the autonomous driving vehicle 60 may further include an input device 603 and an output device 604 , which are interconnected via a bus system and / or other forms of connection mechanisms (not shown).
[0114] The input device 603 may include, for example, a keyboard, a mouse, and the like.
[0115] The output device 604 can output various information to the outside, including characteristic information of the target to be detected, the first tracking trajectory, the second tracking trajectory, etc. The output device 604 can include, for example, a display, a speaker, a printer, a communication network and its connected remote output device, etc.
[0116] Of course, to simplify, Figure 6 Only some of the components related to the present application in the autonomous driving vehicle 60 are shown, and components such as buses, input / output interfaces, etc. are omitted. In addition, the autonomous driving vehicle 60 may also include any other appropriate components depending on the specific application.
[0117] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the target tracking method according to various embodiments of the present application described above in this specification.
[0118] The computer program product may be written in any combination of one or more programming languages to implement the program code for performing the operations of the embodiments of the present application, including object-oriented programming languages such as Java, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.
[0119] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enables the processor to execute the steps of the target tracking method according to various embodiments of the present application described above in this specification.
[0120] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.
[0121] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this application are merely illustrative and not restrictive, and it should not be assumed that these advantages, strengths, and effects are required of each embodiment of this application. In addition, the specific details disclosed above are merely illustrative and facilitating understanding, and are not restrictive. The above details do not limit this application to necessarily being implemented using the above specific details.
[0122] The block diagrams of the devices, devices, equipment, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "include," "comprise," "have," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.
[0123] It should also be noted that in the apparatus, device, and method of the present application, each component or each step can be decomposed and / or recombined, and such decomposition and / or recombination should be regarded as equivalent solutions of the present application.
[0124] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.
[0125] The above description has been provided for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.
Claims
1. A target tracking method, characterized in that: include: Obtaining feature information of the target to be detected contained in multiple frames of environmental data; Based on the feature information of the target to be detected contained in the multiple frames of environmental data, performing time-series correlation on the same target to be detected in different frames of environmental data to obtain a first tracking trajectory of the same target to be detected; If the first tracking trajectory of the same target to be detected meets the trajectory screening condition, a trajectory prediction is performed on the same target to be detected to obtain a second tracking trajectory of the same target to be detected, wherein the trajectory screening condition includes at least one of the following items: the first tracking trajectory cannot cover the time series range of all the multiple frames of environmental data; there are at least two first tracking trajectories of the same target to be detected.
2. The target tracking method according to claim 1, characterized in that The step of performing temporal correlation on the same target to be detected in different frames of environmental data based on the feature information of the target to be detected contained in the multiple frames of environmental data to obtain a first tracking trajectory of the same target to be detected includes: Based on the feature information of the target to be detected contained in the multiple frames of environmental data, construct an initial tracking trajectory corresponding to the same target to be detected and including a time sequence identifier; If there is a timing interruption in the initial tracking trajectory, the initial tracking trajectory is interpolated and completed to obtain a first tracking trajectory of the same target to be detected.
3. The target tracking method according to claim 2, characterized in that The feature information of the target to be detected includes an instance identifier and a detection frame of the target to be detected; and constructing an initial tracking trajectory corresponding to the same target to be detected and including a timing identifier based on the feature information of the target to be detected contained in the multiple frames of environmental data includes: Based on the instance identifier and detection frame of the target to be detected in each frame of environmental data, the trace formed by the detection frame with the same instance identifier is used as the initial tracking trajectory of the same target to be detected; Time-sorting the multiple frames of environmental data to obtain a time sequence number for each frame of environmental data; The time sequence number of the environment data where each detection frame on the initial tracking trajectory is located is recorded to obtain the initial tracking trajectory corresponding to the same target to be detected and including the time sequence identifier.
4. The target tracking method according to claim 3, characterized in that The method further comprises: If the same target to be detected corresponds to multiple initial tracking trajectories, and there is temporal overlap between the multiple initial tracking trajectories, the multiple initial tracking trajectories of the same target to be detected are merged to obtain only one initial tracking trajectory of the same target to be detected.
5. The target tracking method according to any one of claims 1 to 3, characterized in that: The step of performing trajectory prediction on the same target to be detected to obtain a second tracking trajectory of the same target to be detected includes: Predicting a trajectory position of the same target to be detected at a timing interruption based on the first tracking trajectory of the same target to be detected; Based on the characteristic information of the target to be detected, the trajectory position of the same target to be detected at the time sequence interruption is matched with the trajectory positions detected at other time sequences; The successfully matched trajectory position is fused with the first tracking trajectory to generate a second tracking trajectory of the same target to be detected.
6. The target tracking method according to any one of claims 1 to 3, characterized in that: The characteristic information of the target to be detected includes a detection frame of the target to be detected; before acquiring the characteristic information of the target to be detected contained in the multi-frame environmental data, the method further includes: Acquiring initial feature information of the target to be detected contained in the multiple frames of environmental data; Perform at least one of the following filtering on the initial feature information of the target to be detected to obtain the feature information of the target to be detected: Performing spatial filtering on the detection frame of the target to be detected, and retaining the detection frame located within the predefined area; Based on a preset confidence threshold, confidence filtering is performed on the detection frame of the target to be detected; Based on a preset overlap rate threshold, the detection frame of the target to be detected is filtered for overlapping areas.
7. The target tracking method according to claim 6, characterized in that: The predefined area includes an area with a target distance range centered on the vehicle; and / or, The confidence threshold is set based on the category of the target to be detected and / or the distance between the detection frame of the target to be detected and the vehicle; and / or, The overlap rate threshold is set based on the category of the target to be detected.
8. The target tracking method according to claim 6, characterized in that: The multi-frame environmental data includes multi-frame image data and multi-frame point cloud data; the obtaining of initial feature information of the target to be detected contained in the multi-frame environmental data includes: The multi-frame image data and the multi-frame point cloud data are input into a multimodal sparse query network to obtain initial feature information of the target to be detected, wherein the multimodal sparse query network is obtained by supervised training on instance identification of multimodal data.
9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and the computer program is used to execute the target tracking method according to any one of claims 1 to 8.
10. An autonomous driving vehicle, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is used to execute the target tracking method described in any one of claims 1 to 8.