Spatial point cloud-based regional target tracking method and system, and computer device
By placing information in the recovery area when the target disappears and using point cloud data for repair, the problem of target loss after the user stops is solved, improving the accuracy of target tracking and the success rate of cascading operations.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- ULTIMATE IOT (SHANGHAI) TECH LTD
- Filing Date
- 2025-10-30
- Publication Date
- 2026-07-23
AI Technical Summary
In existing technologies, if a user remains stationary or makes only small movements after entering an area, target detection is easily lost, leading to the failure of cascaded operations.
By acquiring point cloud data of the detection area, if the target disappears, its information is placed in a preset recovery area, and the target information is repaired using micro-dynamics or dynamic point cloud within a preset time, and then it is put back into the target tracker for tracking.
It reduces the occurrence of target loss, decreases the occurrence of misjudgment and cascading operation failures, and improves the accuracy of target tracking.
Smart Images
Figure CN2025131341_23072026_PF_FP_ABST
Abstract
Description
Regional target tracking methods, systems, and computer equipment based on spatial point clouds
[0001] Cross-references to related applications
[0002] This application claims priority to Chinese Patent Application No. 202510073415X, filed on January 17, 2025, entitled “Regional Target Tracking Method, System and Computer Device Based on Spatial Point Cloud”, the entire contents of which are incorporated herein by reference. Technical Field
[0003] This disclosure relates to a method, system, and computer equipment for regional target tracking based on spatial point clouds, and belongs to the field of smart home. Background Technology
[0004] With the rapid development of IoT technology, smart home devices are playing an increasingly important role in users' lives. Different home devices can also be linked together. For example, when a target (i.e., a user) is detected in the detection area, home devices can be linked to trigger cascading events, such as turning on lights and air conditioners. When no target is detected in the detection area, the cascading events end, such as turning off lights and air conditioners.
[0005] However, existing methods can easily cause the tracking target (i.e., the user) to disappear when the user is stationary or moves very little after entering the detection area, resulting in misjudgment. That is, a message indicating that no one is present will be given when someone is present, causing the cascading operation to fail. Summary of the Invention
[0006] The purpose of this disclosure is to provide a regional target tracking method, system, and computer device based on spatial point clouds, in order to solve the problem in the prior art that target detection is easily lost after a person enters the area and stops, leading to the failure of cascading operations.
[0007] To achieve the above objectives, the disclosed solution includes:
[0008] This disclosure discloses a method for regional target tracking based on spatial point clouds, comprising the following steps:
[0009] Acquire point cloud data of the detection area;
[0010] If a strong target disappears from the detection area tracked by the target tracker in the current frame point cloud data, the information of the strong target tracked before the corresponding time in the current frame will be placed in the preset recovery area.
[0011] If any frame of point cloud data contains a first preset number of micro-dynamic point clouds or a second preset number of dynamic point clouds within a preset recovery time after the current frame's corresponding time, the information of the strong target in the recovery area is used to repair the information of the strong target in the missing time period and then the information of the strong target is put back into the target tracker to continue tracking the strong target.
[0012] Strong targets are targets that require close attention; micro-dynamic point clouds represent non-static point clouds with relatively small movement amplitudes; dynamic point clouds represent non-static point clouds other than micro-dynamic point clouds; the second preset number is less than the first preset number; the missing time period represents the time period between the current frame and the frame corresponding to the appearance of the first preset number of micro-dynamic point clouds or the second preset number of dynamic point clouds; the target tracker is configured to perform target tracking.
[0013] Optionally, methods for confirming whether a target is a strong target include:
[0014] Obtain trajectory information of each target tracked in the target tracker;
[0015] If the trajectory information of any target indicates that the target's trajectory enters the detection area from outside the detection area, and the trajectory position data within the detection area is greater than a preset number, and the point cloud data density of each frame of any target is greater than a preset density, then any target is labeled as a strong target to characterize any target as a strong target.
[0016] Optionally, after assigning a strong target label to any target, the strong target label is updated;
[0017] The process of updating strong target labels includes deleting the strong target label when any target moves out of the detection area for at least one frame.
[0018] Optionally, the process of repairing the information of strong targets in the missing time period using the information of strong targets in the recovery area includes: updating the information of strong targets in the missing time period to the information of strong targets in the previous frame of the current frame.
[0019] Optionally, the target tracking process of the target tracker includes: acquiring each target to be matched in the point cloud data of the frame to be matched, matching each target to be matched with each target in the target tracker respectively, and if any target to be matched and any target in the target tracker satisfy all the matching conditions, then the two are considered to be successfully matched.
[0020] The matching conditions include: the straight-line distance on the horizontal plane between the centroid position of any target to be matched and the latest position of any target in the target tracker is less than a preset straight-line distance.
[0021] Optionally, the matching conditions also include: the distances between the centroid position of any target to be matched and the latest position of any target in the target tracker on the two mutually perpendicular components in the horizontal plane respectively satisfy the corresponding preset supplementary distances.
[0022] Optionally, the targets to be matched are divided into primary targets and secondary targets. When matching primary targets and secondary targets with each target in the target tracker, primary targets are first matched with each target in the target tracker. After the primary targets are matched, secondary targets are matched with targets in the target tracker other than those that were successfully matched with primary targets.
[0023] The preset straight-line distance corresponding to the first-level target is greater than the preset straight-line distance corresponding to the second-level target; the first-level target refers to the target to be matched that contains more dynamic point clouds than the third preset number or more non-static point clouds than the fourth preset number; the second-level target refers to the target to be matched that contains less dynamic point clouds than the third preset number or less non-static point clouds than the fourth preset number.
[0024] Optionally, dividing the target to be matched into primary targets and secondary targets includes:
[0025] All the point cloud data of the frame to be matched in the millimeter-wave radar coordinate system are converted into point cloud data in the world coordinate system.
[0026] Point cloud filtering is performed on the point cloud data in the world coordinate system.
[0027] After filtering, based on the frequency Doppler effect, the non-static point cloud data is distinguished into dynamic point clouds and micro-dynamic point clouds.
[0028] The dynamic point cloud is clustered, and then the scattered points are filtered out.
[0029] After filtering out scattered points, the target is evaluated based on the preset distance between each cloud cluster and the preset tracking target to determine whether the target is the target to be tracked.
[0030] The number of dynamic point clouds of the tracked target is determined by analyzing the density and voxel density of the dynamic point cloud after filtering out scattered points.
[0031] Based on the number of dynamic point clouds or non-static point clouds of the tracked target, the target to be matched is divided into the primary target and the secondary target.
[0032] Optionally, the primary target is first matched with each target in the target tracker, including:
[0033] If the straight-line distance on the horizontal plane between the centroid position of any primary target to be matched and the latest position of any target in the target tracker is less than a first preset straight-line distance, then it is determined that the primary target to be matched and any target in the target tracker are successfully matched.
[0034] Optionally, the secondary target is matched with targets in the target tracker that are not matched with the primary target, including:
[0035] If the straight-line distance on the horizontal plane between the centroid position of any secondary target to be matched and the latest position of any target in the target tracker other than the target that successfully matched the primary target is less than a second preset straight-line distance, then it is determined that the secondary target to be matched and the target in the target tracker are successfully matched.
[0036] Optionally, the target information includes: index number, whether it is a strong target label, and whether it is a strong motion label;
[0037] The index number is the label of each target;
[0038] When the position difference between any target in any two consecutive frames in any direction on the horizontal plane is greater than a preset position difference, the target is labeled as a strong motion target.
[0039] Optionally, the detection area is the area excluding the interference area; the interference area is the area that does not need to be detected.
[0040] Optionally, the target tracker's target tracking process also includes: filtering out symmetrical mirror targets;
[0041] A symmetrical mirror target is a target in two consecutive frames of point cloud data where the velocity components in two mutually perpendicular directions on the horizontal plane are equal.
[0042] Optionally, the regional target tracking method based on spatial point clouds further includes: outputting whether there is a target in the detection area, the number of targets, and the location of the targets based on the tracking results of the target tracker.
[0043] Optionally, before the target tracker performs target tracking, the method further includes:
[0044] The target tracker is initialized based on the detection area, folding area, rotation angle, and the number of times the target is set as a fixed target.
[0045] The initialization includes at least: initialization of target data classification, initialization of single-target tracking, and initialization of multi-target set update.
[0046] The present disclosure discloses a computer device including a processor, characterized in that the processor executes a computer program to implement the steps of the regional target tracking method based on spatial point clouds as described.
[0047] This disclosure discloses a regional target tracking system based on spatial point clouds, including a radar and a processor. The radar is configured to acquire point cloud data in a detection area and transmit the point cloud data to the processor. The processor is configured to execute a computer program to implement the steps of the regional target tracking method based on spatial point clouds as described above.
[0048] The beneficial effects of this disclosure are as follows: As a pioneering invention, the regional target tracking method, system, and computer device based on spatial point clouds provided by this disclosure can determine the number of targets in the detection area by acquiring point cloud data of the detection area; by observing the disappearance of strong targets within the detection area tracked by the target tracker in the current frame point cloud data, the information of the strong targets tracked before the corresponding time of the current frame is placed in a preset recovery area, which can extend the time of strong target disappearance and reduce the occurrence of misjudgment due to misidentification; if a first preset number of micro-dynamic point clouds or a second preset number of dynamic point clouds appear in any frame point cloud data within a preset recovery time after the corresponding time of the current frame, the information of the strong targets in the recovery area is used to repair the information of the strong targets in the missing time period, and then the information of the strong targets is put back into the target tracker to continue tracking the strong targets. By repairing the trajectory of the strong targets, the occurrence of trajectory loss leading to tracking failure can be reduced. This disclosure can reduce the occurrence of target loss, thereby reducing the problem of cascading operation failure. Attached Figure Description
[0049] Figure 1 is a flowchart illustrating a regional target tracking method based on spatial point clouds provided in an embodiment of this disclosure;
[0050] Figure 2 is a schematic diagram of a trajectory update process provided in an embodiment of this disclosure. Detailed Implementation
[0051] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.
[0052] The concept of this disclosure is to improve the resistance to the risk of target detection being easily lost when a person is stationary, which could lead to the failure of cascaded operations, by setting a short-term recovery area where the target disappears.
[0053] Specifically, by acquiring point cloud data of the detection area, the number of targets in the detection area can be determined. If a strong target disappears from the detection area tracked by the target tracker in the current frame's point cloud data, the information of the strong targets tracked before the corresponding time in the current frame is placed in a preset recovery area. This extends the time before the strong target disappears, reducing the occurrence of misjudgments due to misidentification. If, within a preset recovery time after the corresponding time in the current frame, a first preset number of micro-dynamic point clouds or a second preset number of dynamic point clouds appear in any frame's point cloud data, the information of the strong targets in the recovery area is used to repair the information of the strong targets during the missing time period, and then the information of the strong targets is put back into the target tracker to continue tracking the strong targets. By repairing the trajectory of the strong targets, the occurrence of trajectory loss leading to tracking failure can be reduced. This disclosure can reduce the occurrence of target loss, thereby reducing the problem of cascading operation failure.
[0054] An embodiment of a region target tracking method based on spatial point clouds:
[0055] Figure 1 is a flowchart of a regional target tracking method based on spatial point cloud provided in an embodiment of this disclosure. As shown in Figure 1, the method includes the following steps: S101, acquiring point cloud data of the detection area.
[0056] Specifically, point cloud data of the detection area is acquired according to a preset sampling frequency and a preset number of sampling frames.
[0057] The preset sampling frequency can be 1 second, 2 seconds, etc., and can be set according to the actual situation. This disclosure does not impose any particular limitation on it. The following explanation uses a preset sampling frequency of 1 second as an example.
[0058] The preset sampling frame count can be 5 frames, 10 frames, etc., and can be set according to the actual situation. This disclosure does not impose any particular limitation on it. The following explanation uses a preset sampling frame count of 5 frames as an example.
[0059] When the preset sampling frequency is 1 second and the preset sampling frame count is 5 frames: the point cloud data of the detection area is acquired according to the acquisition method of 5 frames per second.
[0060] The detection area is the area excluding the interference area; the interference area is the area that does not need to be detected.
[0061] S102. If a strong target disappears from the detection area tracked by the target tracker in the current frame point cloud data, the information of the strong target tracked before the corresponding time in the current frame is placed in the preset recovery area.
[0062] The methods for determining whether a target is a strong target include the following steps S102A and S102B.
[0063] S102A: Obtain trajectory information of each target tracked in the target tracker.
[0064] The trajectory information can be obtained from 100 frames, or other numbers. This disclosure does not impose any particular limitation on this. The following example of obtaining 100 frames will be used for illustrative purposes.
[0065] Specifically, if the trajectory information of the target exceeds 100 frames, only the most recent 100 frames of trajectory information are taken; if the trajectory data of the target is less than 100 frames, all frames of trajectory information are read.
[0066] S102B: When the trajectory information of any target indicates that the trajectory of the target enters the detection area from outside the detection area, and the trajectory position data in the detection area is greater than a preset number, and the point cloud data density of each frame of any target is greater than a preset density, then a strong target label is applied to any target to characterize any target as a strong target.
[0067] To improve tracking efficiency, after assigning a strong target label to any target, the strong target label needs to be updated. The process of updating the strong target label includes deleting the strong target label when any target moves out of the detection area for at least one frame.
[0068] Among them, strong targets are those that require special attention; the disappearance of a strong target can be the disappearance of its label or the disappearance of the strong target itself, and this disclosure does not make any special restrictions on this.
[0069] It is understandable that whether the label of a strong target disappears or the strong target itself disappears, a large number of point clouds will disappear or shrink. After this phenomenon occurs, placing the information of the strong target tracked before the corresponding time of the current frame into the preset recovery area can prolong the time of strong target disappearance and reduce the occurrence of misjudgment due to misidentification.
[0070] S103. If, within a preset recovery time after the current frame's corresponding time, any frame of point cloud data contains a first preset number of micro-dynamic point clouds or a second preset number of dynamic point clouds, then the information of the strong target in the recovery area is used to repair the information of the strong target in the missing time period, and then the information of the strong target is put back into the target tracker to continue tracking the strong target.
[0071] Among them, micro-dynamic point cloud refers to non-static point cloud with relatively small movement amplitude; dynamic point cloud refers to non-static point cloud other than micro-dynamic point cloud; the second preset number is less than the first preset number; the missing time period refers to the time period between the moment when the strong target disappears and the moment when a frame corresponding to the first preset number of micro-dynamic point clouds or the second preset number of dynamic point clouds appears; the target tracker is configured to perform target tracking; the preset recovery time can be 2 minutes or other times, etc., which can be selected according to the actual situation, and this disclosure does not make any special limitation on this.
[0072] The first preset number of micro-dynamic point clouds or the second preset number of dynamic point clouds may be disappearing strong targets or new targets, etc., and this disclosure does not make any special restrictions on this.
[0073] The process of repairing the information of strong targets in the missing time period using information from strong targets in the recovery area includes updating the information of the strong targets in the missing time period with the information of the strong targets in the previous frame of the current frame. Repairing the trajectory of strong targets can reduce the occurrence of tracking failures due to trajectory loss.
[0074] To improve tracking efficiency, as an optional implementation method, the target tracking process in the target tracker may also include: step S104, obtaining each target to be matched in the point cloud data of the frame to be matched, matching each target to be matched with each target in the target tracker, and if any target to be matched satisfies all the matching conditions with any target in the target tracker, then the two are considered to be successfully matched.
[0075] The matching conditions include: the straight-line distance on the horizontal plane between the centroid position of any target to be matched and the latest position of any target in the target tracker is less than a preset straight-line distance.
[0076] To reduce the occurrence of false matches, as an optional implementation, the matching condition further includes: the distance between the centroid position of any target to be matched and the latest position of any target in the target tracker on the two mutually perpendicular components in the horizontal plane respectively satisfies the corresponding preset supplementary distance.
[0077] The targets to be matched can be divided into primary targets and secondary targets.
[0078] The process of dividing primary and secondary objectives may include the following sub-steps: step S104A and step S104B.
[0079] S104A: Initialize the acquired point cloud data.
[0080] Specifically, S104A1 converts all point clouds in the millimeter-wave radar coordinate system into point clouds in the real-world coordinate system.
[0081] Specifically, the rotation matrix is established based on the angle of the millimeter-wave radar chip: after testing, it was found that the point cloud effect in the plane is best when the angle is 45 degrees, and the rotation matrix at this time is T; traverse all the point cloud coordinates, and multiply each point cloud by the rotation matrix T to obtain the coordinate position in the real world.
[0082] S104A2, perform point cloud filtering on point cloud data in the world coordinate system.
[0083] Specifically, height and low-degree thresholds are obtained through scene statistical analysis. All points in each point cloud are then traversed, and all high-degree and low-degree interference points are filtered out using these thresholds. The purpose of filtering out these interference points is to ensure that the resulting point cloud is as close as possible to the target point cloud.
[0084] S104A3. After filtering, based on the frequency Doppler effect, non-static point cloud data is distinguished into dynamic point clouds and micro-dynamic point clouds.
[0085] S104A4. Cluster the dynamic point cloud and evaluate whether the target is the one that truly needs to be tracked based on the clustering results.
[0086] The point cloud data of the detection area obtained directly may include point cloud data other than users. This point cloud data does not need to be tracked and is a false target. Therefore, it is necessary to cluster the dynamic point cloud to obtain the real target, that is, the target that needs to be tracked.
[0087] The clustering can be performed using a density-based spatial clustering of applications with noise (DBSCAN) algorithm, or other clustering algorithms; this disclosure does not impose any particular limitation on this. The following explanation will use the DBSCAN algorithm as an example.
[0088] For the DBSCAN algorithm, specifically, 1) calculate the ε-neighborhood of all points: For each point P in the dataset, calculate how many neighbors it has in its ε-neighborhood. The threshold for this number of neighbors is usually defined by a parameter MinPts.
[0089] 2) Marking core points: If the number of points in the ε-neighborhood of a point is greater than or equal to MinPts, then this point is marked as a core point.
[0090] 3) Find density-connected points: For each core point, find all points density-connected to it. If point P is in the ε-neighborhood of point O, and O is a core point, then P is a point density-connected to O.
[0091] 4) Marking noise points and boundary points: Points not marked as core points are marked as noise points. Points connected to any core point density but not themselves are marked as boundary points.
[0092] 5) Assign a unique cluster label to each core point or point density-connected to it: Assign a unique cluster label to each core point or point density-connected to it. If a point is density-connected to multiple core points, it will be assigned the cluster label of the first found core point.
[0093] S104A5. After clustering, the scattered points are filtered out.
[0094] After clustering, there may be some scattered points that will reduce the effectiveness of subsequent tracking. Therefore, these scattered points need to be filtered out.
[0095] Specifically, noisy point cloud clusters in the point cloud data are removed, as are point cloud clusters in the point cloud data whose number of points is less than the first preset number.
[0096] Among them, the noise point cloud cluster is a point cloud cluster that does not belong to the target to be tracked, such as a table, chair, air conditioner, etc.; the first preset number of points can be 10 points, or 5 points, etc. This disclosure does not make any special limitation on this. The following will take the first preset number of points of 10 points as an example for illustrative explanation.
[0097] As one optional implementation, noisy point cloud clusters can be removed first, followed by point cloud clusters smaller than a first preset number of points. As another optional implementation, point cloud clusters smaller than a first preset number of points can be removed first, followed by noisy point cloud clusters. This disclosure does not impose any particular limitation on this method; the following explanation will use the example of removing noisy point cloud clusters first, followed by removing point cloud clusters smaller than a first preset number of points.
[0098] S104A6. After filtering out scattered points, assess whether the target is the real target that needs to be tracked based on the preset distance between each point cloud cluster and the preset tracking target.
[0099] For example, suppose there are 5 clusters of point clouds, one of which has fewer than 10 points; suppose the preset tracking targets include two preset tracking targets, a first preset tracking target and a second preset tracking target.
[0100] Point cloud clusters with fewer than 10 points are removed, leaving 4 clusters. The distances between these 4 clusters and the first preset tracking target, and the distances between these clusters and the second preset tracking target, are calculated. Clusters matching the first preset distance are assigned to the first preset tracking target, and clusters matching the second preset distance are assigned to the second preset tracking target. After classifying the point cloud clusters, target tracking is performed based on the classification results.
[0101] S104A7. Analyze the dynamic point cloud after filtering out scattered points to determine the number of points in the target's dynamic point cloud based on its density and voxel density. Voxel density refers to the number of points in a unit space.
[0102] S104B. Determine the target type of the target to be matched based on the number of dynamic point clouds or the number of non-static point clouds.
[0103] Specifically, a target to be matched that contains more than a third preset number of dynamic point clouds or more than a fourth preset number of non-static point clouds is a first-level target; a target to be matched that contains less than a third preset number of dynamic point clouds or less than a fourth preset number of non-static point clouds is a second-level target.
[0104] The third preset number can be 5 or other values, which can be set according to the actual situation. This disclosure does not make any special limitation on this. The following explanation will take the third preset number of 5 as an example. The fourth preset number can be 10 or other values, which can be set according to the actual situation. This disclosure does not make any special limitation on this. The following explanation will take the fourth preset number of 10 as an example.
[0105] When dividing the targets to be matched into primary targets and secondary targets, the specific matching process is as follows:
[0106] If the straight-line distance on the horizontal plane between the centroid of any primary target to be matched and the latest position of any target in the target tracker is less than a first preset straight-line distance, then the two are successfully matched.
[0107] A match is successful when the straight-line distance on the horizontal plane between the centroid position of any secondary target to be matched and the latest position of any target in the target tracker (excluding targets that have successfully matched with primary targets) is less than a second preset straight-line distance. The second preset straight-line distance is less than the first preset straight-line distance. The first preset straight-line distance can be 0.8 meters or other numbers; this disclosure does not impose any particular limitation on it. The following explanation uses 0.8 meters as an example. The second preset straight-line distance can be 0.5 meters or other numbers; this disclosure does not impose any particular limitation on it. The following explanation uses 0.5 meters as an example.
[0108] To reduce the occurrence of false matches, as another optional implementation, the straight-line distance on the horizontal plane between the centroid position of any primary target to be matched and the latest position of any target in the target tracker is less than a first preset straight-line distance, and the distances between the two on the two vertical components of the horizontal plane respectively satisfy the corresponding preset supplementary distances, then the two are successfully matched.
[0109] A match is successful when the straight-line distance between the centroid position of any secondary target to be matched and the latest position of any target in the target tracker other than the target that successfully matched the primary target is less than the second preset straight-line distance on the horizontal plane, and the distances between the two targets on the two vertical components of the horizontal plane satisfy the corresponding preset supplementary distances.
[0110] To improve matching efficiency, as an optional implementation method, the trajectories that have been successfully matched with the primary target can be erased before matching the secondary target.
[0111] The corresponding preset supplementary distance is either a horizontal distance or a vertical distance. The value of the preset supplementary distance can be 0.45 meters or other values, which can be set according to actual needs. This disclosure does not impose any special limitations on this. This embodiment uses 0.45 meters as an example for illustrative purposes.
[0112] S105. Update the information of the first-level or second-level target that has been successfully matched with the target in the target tracker to the corresponding successfully matched target.
[0113] The target information can include the target's speed, acceleration, trajectory information, whether it is a strong target label, and whether it is a strong motion label, etc.
[0114] Among them, the index number is the label of each target; strong targets are targets that need to be focused on; when the position difference between any target in any direction on the horizontal plane is greater than the preset position difference in two consecutive frames, the strong motion label is applied to the target to indicate that the target is a strong motion target.
[0115] To improve tracking accuracy, as an optional implementation, the target tracking process in the target tracker may also include: step S105, filtering out symmetrical mirror targets.
[0116] Among them, the symmetrical mirror target is the target in two consecutive frames of point cloud data where the velocity components in the two vertical directions corresponding to the horizontal plane are equal.
[0117] Specifically, select data from two consecutive frames and check whether the velocity components of the X-axis and Y-axis of the same target are the same in the previous and next frames. If they are the same, check whether it is symmetrical about any point or any plane. If it is symmetrical about any point or any plane, then the target is a symmetrical mirror target.
[0118] S106. Based on the tracking status of the target tracker, output whether there is a target in the detection area, the number of targets, and the position of the targets, and update the target tracker according to the current frame state.
[0119] The number of targets can be one or more, and this disclosure does not impose any particular limitation on this.
[0120] Figure 2 is a schematic flowchart of a trajectory update provided in an embodiment of this disclosure. As shown in Figure 2, there are multiple target detections, multiple old trajectories within the region, and multiple new trajectories within the region. In the first matching of old trajectories and targets based on spatial distance, some targets are matched with old trajectories, while others are not. A second matching is performed on the targets that were not matched with old trajectories, the old trajectories that were not matched with targets (i.e., the remaining old trajectories and the remaining targets), all new trajectories, and new targets, based on spatial distance. During this second matching, some targets (including the remaining old targets and new targets) are matched with new trajectories, while others are matched with old trajectories.
[0121] Delete the target to be tested that has not been matched with a trajectory twice in a row (i.e., unmatched detection), delete the old trajectory that has not been matched with a target to be tested within a preset time, delete the remaining old target to be tested that has not been matched with a trajectory twice in a row, and temporarily designate the new trajectory that has not been matched with any target as a new trajectory.
[0122] Before the target tracker performs target tracking, step S107 can be executed to initialize the multi-target tracker. Specifically, the tracker is first initialized based on relevant parameters such as the detection area, folding area, rotation angle, and the number of times the target is set as a fixed target. Initialization includes at least: initialization of target data classification, initialization of single-target tracking, and initialization of multi-target set update.
[0123] Among them, the detection area refers to the area in the room that needs to be detected; the folded area refers to the area in the room that is not detected, i.e., the interference area; and the rotation angle is the rotation angle required to convert the point cloud in the millimeter-wave radar coordinate system to the point cloud in the real coordinate system.
[0124] The initialization of target data classification refers to the initialization process of the collected point cloud data in the multi-target tracker, which divides the point cloud into dynamic point cloud and micro-dynamic point cloud according to the number of points and distances in the point cloud.
[0125] The initialization of single-target tracking refers to setting how many consecutive frames a target needs to be detected to be considered a true target, as well as some information about the target itself. For example, if 5 frames of data are collected per second, and the target appears in 5 consecutive frames per second, it is considered a true target and is added to the multi-target tracker as a preset target; if the target does not appear in 5 consecutive frames per second, it is not considered a true target and is not added to the multi-target tracker.
[0126] Initialization of multi-objective set updates refers to the fact that the multi-objective aggregator also needs initialization parameter settings; the multi-objective aggregator is used to store information related to multiple objectives. For example, if a strong objective disappears while moving, this objective is stored and retrieved when it reappears, allowing for faster initialization.
[0127] This disclosure provides a method for regional target tracking based on spatial point clouds. By acquiring point cloud data of a detection area, the number of targets in the detection area can be determined. If a strong target disappears from the detection area tracked by the target tracker in the current frame's point cloud data, the information of the strong targets tracked before the corresponding time in the current frame is placed in a preset recovery area. This extends the time before the strong target disappears, reducing the occurrence of misjudgments due to misidentification. If a first preset number of micro-dynamic point clouds or a second preset number of dynamic point clouds appear in any frame of point cloud data within a preset recovery time after the corresponding time in the current frame, the information of the strong targets in the recovery area is used to repair the information of the strong targets in the missing time period, and then the information of the strong targets is put back into the target tracker to continue tracking the strong targets. By repairing the trajectory of the strong targets, the occurrence of trajectory loss leading to tracking failure can be reduced. This disclosure can reduce the occurrence of target loss, thereby reducing the problem of cascading operation failure.
[0128] An embodiment of a computer device:
[0129] The present disclosure includes a computer device comprising a processor that executes a computer program to implement the steps of the above-described method for tracking regional targets based on spatial point clouds.
[0130] This disclosure provides a computer device that can achieve the same beneficial effects as the aforementioned method for tracking regional targets based on spatial point clouds, which will not be repeated here.
[0131] An embodiment of a regional target tracking system based on spatial point clouds:
[0132] This disclosure provides a regional target tracking system based on spatial point clouds, including a radar and a processor. The radar is configured to acquire point cloud data in a detection area and transmit the point cloud data to the processor. The processor is configured to execute a computer program to implement the steps of the regional target tracking method based on spatial point clouds as described above.
[0133] The embodiments disclosed herein provide a regional target tracking system based on spatial point clouds that can achieve the same beneficial effects as the aforementioned regional target tracking method based on spatial point clouds, so it will not be described again here. Industrial applicability
[0134] In this embodiment, if a strong target disappears from the detection area tracked by the target tracker in the current frame point cloud data, the information of the strong target tracked before the corresponding time of the current frame is placed in a preset recovery area. If a first preset number of micro-dynamic point clouds or a second preset number of dynamic point clouds appear in any frame point cloud data within a preset recovery time after the corresponding time of the current frame, the information of the strong target in the recovery area is used to repair the information of the strong target in the missing time period, and then the information of the strong target is put back into the target tracker to continue tracking the strong target. This can reduce the occurrence of target loss, thereby reducing the occurrence of cascading operation failures and improving practicality.
Claims
1. A method for tracking a region target based on a spatial point cloud, characterized in that, Includes the following steps: Acquire point cloud data of the detection area; If a strong target disappears from the detection area tracked by the target tracker in the current frame point cloud data, the information of the strong target tracked before the corresponding time of the current frame is placed in a preset recovery area; If, within a preset recovery time after the current frame, any frame of point cloud data contains a first preset number of micro-dynamic point clouds or a second preset number of dynamic point clouds, then the information of the strong target in the recovery area is used to repair the information of the strong target in the missing time period, and then the information of the strong target is put back into the target tracker to continue tracking the strong target. The strong target is the target that needs to be focused on; the micro-dynamic point cloud refers to a non-static point cloud with a relatively small movement amplitude; the dynamic point cloud refers to a non-static point cloud other than the micro-dynamic point cloud; the second preset number is less than the first preset number; the missing time period refers to the time period between the moment corresponding to the current frame and the moment corresponding to a frame in which the first preset number of micro-dynamic point clouds or the second preset number of dynamic point clouds appear; the target tracker is configured to perform target tracking.
2. The spatial point cloud based region object tracking method of claim 1, wherein, Methods for confirming whether something is a strong target include: Obtain the trajectory information of each target tracked by the target tracker; If the trajectory information of any target indicates that the trajectory of any target enters the detection area from outside the detection area, and the trajectory position data in the detection area is greater than a preset number, and the point cloud data density of each frame of any target is greater than a preset density, then a strong target label is assigned to any target to characterize any target as a strong target.
3. The spatial point cloud based region object tracking method of claim 2, wherein, After assigning the strong target label to any of the targets, the strong target label is updated; The process of updating the strong target label includes: deleting the strong target label when any target moves out of the detection area for at least one frame.
4. The spatial point cloud based area object tracking method according to any one of claims 1-3, characterized in that, The process of repairing the information of the strong target in the missing time period using the information of the strong target in the recovery area includes: updating the information of the strong target in the missing time period to the information of the strong target in the previous frame of the current frame.
5. The spatial point cloud based area object tracking method according to any one of claims 1-4, characterized in that, The target tracker performs target tracking as follows: acquiring each target to be matched in the point cloud data of the frame to be matched, matching each target to be matched with each target in the target tracker, and if any target to be matched satisfies all the matching conditions with any target in the target tracker, then the two are considered to be successfully matched. The matching conditions include: the straight-line distance on the horizontal plane between the centroid position of any target to be matched and the latest position of any target in the target tracker is less than a preset straight-line distance.
6. The spatial point cloud based region object tracking method of claim 5, wherein, The matching conditions also include: the distances between the centroid position of any target to be matched and the latest position of any target in the target tracker on the two mutually perpendicular components of the horizontal plane respectively satisfy the corresponding preset supplementary distances.
7. The spatial point cloud based region object tracking method according to claim 5 or 6, characterized in that, The targets to be matched are divided into primary targets and secondary targets. When matching the primary targets and the secondary targets with each target in the target tracker, the primary targets are first matched with each target in the target tracker. After the primary targets are matched, the secondary targets are matched with all targets in the target tracker except those that were successfully matched with the primary targets. The preset straight-line distance corresponding to the first-level target is greater than the preset straight-line distance corresponding to the second-level target; the first-level target refers to a target to be matched that contains more than a third preset number of dynamic point clouds or more than a fourth preset number of non-static point clouds; the second-level target refers to a target to be matched that contains less than a third preset number of dynamic point clouds or less than a fourth preset number of non-static point clouds.
8. The spatial point cloud based region object tracking method of claim 7, wherein, The process of dividing the target to be matched into primary targets and secondary targets includes: All the point cloud data of the frame to be matched in the millimeter-wave radar coordinate system are converted into point cloud data in the world coordinate system. Point cloud filtering is performed on the point cloud data in the world coordinate system. After filtering, based on the frequency Doppler effect, the non-static point cloud data is distinguished into dynamic point clouds and micro-dynamic point clouds. The dynamic point cloud is clustered, and then the scattered points are filtered out. After filtering out scattered points, the target is evaluated based on the preset distance between each cloud cluster and the preset tracking target to determine whether the target is the target to be tracked. The number of dynamic point clouds of the tracked target is determined by analyzing the density and voxel density of the dynamic point cloud after filtering out scattered points. Based on the number of dynamic point clouds or non-static point clouds of the tracked target, the target to be matched is divided into the primary target and the secondary target.
9. The spatial point cloud based region object tracking method according to claim 7 or 8, characterized in that, The primary target is first matched with each target in the target tracker, including: If the straight-line distance on the horizontal plane between the centroid position of any primary target to be matched and the latest position of any target in the target tracker is less than a first preset straight-line distance, then it is determined that the primary target to be matched and any target in the target tracker are successfully matched.
10. The spatial point cloud based area object tracking method according to any one of claims 7 to 9, characterized in that, The secondary target is matched with targets in the target tracker that are not matched with the primary target, including: If the straight-line distance on the horizontal plane between the centroid position of any secondary target to be matched and the latest position of any target in the target tracker other than the target that successfully matched the primary target is less than a second preset straight-line distance, then it is determined that the secondary target to be matched and the target in the target tracker are successfully matched.
11. The spatial point cloud based region object tracking method of claim 3, wherein, The target information includes: index number, whether it is a strong target label, and whether it is a strong motion label; The index number is the label of each target; When the position difference between any two consecutive frames in any direction on the horizontal plane is greater than a preset position difference, the target is labeled with the strong motion tag to characterize the target as a strong motion target.
12. The spatial point cloud based area object tracking method according to any one of claims 1-11, characterized in that, The detection area is the area excluding the interference area; the interference area is the area that does not need to be detected.
13. The spatial point cloud based area object tracking method according to any one of claims 1-12, characterized in that, The target tracker's target tracking process also includes: filtering out symmetrical mirror targets; The symmetrical mirror target is a target in two consecutive frames of point cloud data where the velocity components in two mutually perpendicular directions on the horizontal plane are equal.
14. The spatial point cloud based area object tracking method according to any one of claims 1-13, characterized in that, The regional target tracking method based on spatial point clouds further includes: outputting whether there is a target in the detection area, the number of targets, and the location of the targets based on the tracking status of the target tracker.
15. The spatial point cloud based area object tracking method according to any one of claims 1-14, characterized in that, Before the target tracker performs target tracking, the method further includes: The target tracker is initialized based on the detection area, folding area, rotation angle, and the number of times the target is set as a fixed target. The initialization includes at least: initialization of target data classification, initialization of single-target tracking, and initialization of multi-target set update.
16. A computer device comprising a processor, characterized in that The processor executes a computer program to implement the steps of the regional target tracking method based on spatial point clouds as described in any one of claims 1-15.
17. A spatial point cloud based regional target tracking system comprising a radar and a processor, the radar configured to collect point cloud data in a detection region and transmit the point cloud data to the processor, characterized in that, The processor is configured to execute a computer program to implement the steps of the regional target tracking method based on spatial point clouds as described in any one of claims 1-15.