A target matching method and device

Abstract

This invention discloses a target matching method and apparatus. First, historical targets and predicted targets based on the historical targets at the current time are acquired. Then, target data to be measured is determined based on target point cloud data collected by a 3D sensing device at the current time. For each predicted target of a historical target, a metric parameter is determined based on the target and the predicted target of that historical target to characterize the similarity between the target and the historical target. Finally, the target is matched against historical targets based on the metric parameter relative to each historical target. Thus, the calculation of the metric parameter considers the influence of differences in the disappearance time of historical targets, ensuring that historical targets disappearing within a preset time period can be matched. The influence of shape differences is also considered, with targeted relaxation of the length and width of some targets to be measured, thereby improving the accuracy of target tracking.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a target matching method and apparatus. Background Technology

[0002] When solving 3D multi-target tracking tasks, LiDAR-based detection results differ from image detection. In LiDAR target detection tasks, the target has less feature information in the point cloud, and due to the limitations of point cloud coverage and resolution, the validity (degree of agreement between measurement and reality) of 3D target detection results for features such as category, location, and size is not high.

[0003] Most existing target tracking technologies are designed based on ideal 3D target detection results. Ideally, all targets should be completely detected in every frame of data, and the detection accuracy should be high. However, in actual 3D target detection work, due to factors such as target occlusion within the shooting viewpoint, the quality of point cloud data, point cloud density, or the accuracy of the target detection function, targets may disappear from the target detection results for a period of time before reappearing, or the size validity of the target detection results may be low, thus affecting the target tracking accuracy. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed to provide a target matching method and apparatus that incorporates time difference measurement and shape difference measurement into the calculation of measurement parameters during matching, so as to improve the accuracy of target tracking.

[0005] According to a first aspect of the present invention, a target matching method is provided, comprising:

[0006] Historical targets and predicted targets based on historical targets at the current moment are obtained; historical targets are point cloud targets detected and registered for management within a preset time period before the current moment.

[0007] The target to be measured is determined based on the target point cloud data collected by the 3D sensing device at the current moment;

[0008] For each predicted target of a historical target, based on the target to be tested and the predicted target of that historical target, a metric parameter is determined to characterize the similarity between the target to be tested and that historical target. The metric parameters include a centroid distance metric to characterize the similarity of centroid positions, a temporal difference metric to characterize the continuous disappearance time of the historical target, and a shape difference metric to characterize the length and width similarity between the target to be tested and the historical target. The shape difference metric is determined based on the point cloud density of the target to be tested and the historical target.

[0009] Based on the measurement parameters of the target target relative to each historical target, the target target is matched against the historical targets.

[0010] Optionally, determine the centroid distance metric of the target to be measured relative to the historical target, including:

[0011] Obtain the location information of the first centroid of the predicted target of the historical target, the location information of the second centroid of the target to be tested, and the preset centroid distance matching threshold;

[0012] Based on the center-of-gravity distance matching threshold, the position information of the first center-of-gravity point, and the position information of the second center-of-gravity point, the center-of-gravity distance metric of the target to be tested relative to the historical target is determined.

[0013] Optionally, determine a measure of the time difference between the target and the historical target, including:

[0014] The last moment the historical target appeared, the sampling period of the 3D sensing device, and the preset time difference factor were obtained.

[0015] Based on the current time, the last time the historical target appeared, the sampling period of the 3D sensing device, and the time difference factor, the time difference measure of the target to be measured relative to the historical target is determined.

[0016] Optionally, determine a measure of the shape difference between the target and the historical target, including:

[0017] Obtain the first length, first width, and first point cloud density of the predicted target of the historical target at the last moment of its appearance; obtain the second length, second width, and second point cloud density of the target to be measured, as well as the preset first relaxation factor, preset second relaxation factor, preset shape difference factor, and standard point cloud density.

[0018] Based on the first length, first width, first point cloud density, second length, second width, second point cloud density, first relaxation term factor, second relaxation term factor, shape difference factor, and standard point cloud density, determine the shape difference measure of the target to be measured relative to the historical target.

[0019] Optionally, the metrics may also include point cloud density difference metrics and / or heading difference metrics.

[0020] Optionally, determine a measure of the difference in point cloud density between the target and the historical target, including:

[0021] Obtain the preset point cloud density difference factor, the preset point cloud density difference threshold, the first point cloud density of the historical target at the last moment of its appearance, and the second point cloud density of the target to be measured.

[0022] Based on the point cloud density difference factor, point cloud density difference threshold, first point cloud density, and second point cloud density, the point cloud density difference measure of the target to be measured relative to the historical target is determined.

[0023] Optionally, determine the heading difference metric of the target to be measured relative to the historical target, including:

[0024] Acquire the preset heading difference factor, preset standard angle change rate, preset angle difference threshold, heading angle of the target to be measured, heading angle of the predicted target of the historical target, and the last time the historical target appeared;

[0025] Based on the current time, heading difference factor, standard angle change rate, angle difference threshold, heading angle of the target to be measured, heading angle of the predicted target of the historical target, and the last time the historical target appeared, determine the heading difference measure of the target to be measured relative to the historical target.

[0026] Optionally, the method also includes:

[0027] If a match is successful, the historical target that matches the target to be tested is identified. The target to be tested inherits the ID of the historical target and is added to the trajectory of the historical target.

[0028] If the match fails, a new trajectory and ID will be created for the target to be tested.

[0029] Optionally, based on the target to be tested and the predicted target of the historical target, a metric parameter is determined to characterize the similarity between the target to be tested and the historical target, including:

[0030] Based on the target to be tested and the predicted target of that historical target, the following formula is used to determine the metric parameter used to characterize the similarity between the target to be tested and the historical target:

[0031]

[0032]

[0033] Among them, H ij Let x be the metric parameter of target i relative to historical target j, where i and j are both integers greater than 0; i Let x be the x-axis coordinate of the second centroid of the target i; j The x-axis coordinate of the first centroid of target j is predicted; y i Let y be the y-axis coordinate of the second centroid of the target i; j To predict the y-axis coordinate of the first centroid of target j; l i Let l be the length of the target i to be measured; j To predict the length of target j; w i w is the width of the target i to be measured; j To predict the width of target j; t i t represents the current time; jThe last moment that historical target j appears; s dlmt The centroid distance matching threshold; ρ stac δ is the standard point cloud density; T is the sampling period; δ1 is the time difference factor, δ1≥0; δ4 is the shape difference factor, δ4≥0; γ is the first relaxation term factor, γ≥0; ε is the second relaxation term factor, ε≥0.

[0034] According to a second aspect of the present invention, a target matching device is provided, comprising:

[0035] The acquisition module is used to acquire historical targets and predicted targets based on the current time. Historical targets are point cloud targets detected and registered for management within a preset time period before the current time.

[0036] The determination module is used to determine the target to be measured based on the target point cloud data collected by the 3D sensing device at the current moment;

[0037] The calculation module is used to determine, for each historical target's predicted target, a metric parameter to characterize the similarity between the target to be tested and the predicted historical target, based on the target to be tested and the predicted historical target. The metric parameters include a centroid distance metric to characterize the similarity of the centroid positions, a temporal difference metric to characterize the continuous disappearance time of the historical target, and a shape difference metric to characterize the length and width similarity between the target to be tested and the historical target. The shape difference metric is determined based on the point cloud density of the target to be tested and the historical target.

[0038] The matching module is used to match the target to be tested against historical targets based on the metric parameters of the target to be tested relative to each historical target.

[0039] The above-described one or more technical solutions in the embodiments of this specification have at least the following technical effects:

[0040] The target matching method and apparatus provided in this specification, when matching a target to be tested with historical targets, calculates measurement parameters that take into account the impact of differences in the disappearance time of historical targets. Historical targets that disappear within a preset time period can be matched. An additional time difference measurement is added between the predicted target and the target to be tested for historical targets with longer disappearance times based on the current moment. Furthermore, the influence of shape differences is also considered, i.e., a shape difference measurement is added between the predicted target and the target to be tested for historical targets based on the current moment, including cases where there is a significant shape difference between the target to be tested and the predicted target. During the shape difference measurement calculation, relaxation is applied to small-scale targets and targets with low point cloud density levels based on differences in target detection validity, thereby improving the accuracy of target tracking.

[0041] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0042] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference figures denote the same parts throughout the drawings.

[0043] In the attached diagram:

[0044] Figure 1 A flowchart of a target matching method according to an embodiment of the present invention is shown.

[0045] Figure 2 A block diagram of a target matching device according to an embodiment of the present invention is shown. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0047] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0048] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0049] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0050] This embodiment provides a target matching method, combined with Figure 1 As shown, the target matching method includes steps 101 to 104:

[0051] Step 101: Obtain historical targets and predicted targets based on historical targets at the current moment; historical targets are point cloud targets detected and registered for management within a preset time period before the current moment;

[0052] The target matching method provided in this embodiment can be executed by an electronic device, such as a computer, controller, or server. The 3D sensing device can be a LiDAR (Light Detection and Ranging) system. In one application scenario, a LiDAR system is installed at an intersection to track targets entering the scene. Targets can be vehicles, pedestrians, or other moving objects. This embodiment uses vehicles as an example, and targets can be categorized, such as cars, buses, trucks, and non-motorized vehicles. The 3D sensing device periodically captures images of the intersection and obtains point cloud data. Then, it performs identification and detection based on the point cloud data and records and stores the identified point cloud targets. It should be noted that, with the current time as a reference, point cloud targets detected and identified within a preset time period before the current time are considered historical targets. Point cloud targets detected and identified earlier than the preset time period are not considered in this embodiment. The preset time period can be 3 seconds, 10 seconds, or 30 seconds; this embodiment does not limit this.

[0053] Historical targets detected and identified within a preset time period have corresponding IDs, and their trajectories are recorded and stored, enabling registration and management of historical targets. These historical targets will be accessed and retrieved in subsequent steps. Furthermore, for each historical target, trajectory prediction is performed using the current time as the prediction time, thus obtaining the predicted target for that historical target.

[0054] Step 102: Based on the target point cloud data collected by the 3D sensing device at the current moment, determine the target to be measured;

[0055] In this embodiment, the point cloud data collected by the 3D sensing device at the current moment is used as the target point cloud data. After obtaining the target point cloud data, it is processed, that is, the target point cloud data is identified, mainly identifying target vehicles and / or target pedestrians in the target point cloud data. The identified target vehicles and target pedestrians are the targets to be measured. The targets to be measured identified through the target point cloud data may be one or more, or may be zero.

[0056] It should be noted that these targets may be being captured and identified by the 3D sensing device for the first time; they may have been continuously identified in previously captured point cloud data; or they may have been identified in previously captured point cloud data, but disappeared for a period of time, and then been re-identified in the target point cloud data at the current moment.

[0057] Step 103: For each predicted target of a historical target, determine the measurement parameters used to characterize the similarity between the target to be tested and the predicted target of the historical target, based on the target to be tested and the predicted target of the historical target. The measurement parameters include the centroid distance metric used to characterize the similarity of the centroid positions, the temporal difference metric used to characterize the continuous disappearance time of the historical target, and the shape difference metric used to characterize the length and width similarity between the target to be tested and the historical target. The shape difference metric is determined based on the point cloud density of the target to be tested and the historical target.

[0058] As can be seen from the foregoing, there may be multiple targets to be tested. For each target to be tested, in order to identify whether it is appearing for the first time, this embodiment needs to compare the target to be tested with all historical targets respectively, and calculate the similarity of the target to be tested with respect to all historical targets, so as to determine whether the target to be tested is a historical target or a new target appearing for the first time.

[0059] In this embodiment, for each target to be tested, similarity is calculated by measuring the target's metrics relative to each historical target. Specifically, the metric parameters of the target to be tested relative to the predicted target of each historical target at the current moment are calculated.

[0060] The measurement parameters include a centroid distance metric used to characterize the similarity of centroid positions, a temporal difference metric used to characterize the continuous disappearance time of historical targets, and a shape difference metric used to characterize the length and width similarity between the target to be tested and historical targets. The shape difference metric is determined based on the point cloud density of the target to be tested and historical targets.

[0061] For time difference measurement, it is for targets that appeared before the current moment, but disappeared for a period of time and then reappeared in the target point cloud data. In this embodiment, there is an upper limit to the continuous disappearance time, namely the preset time period mentioned above.

[0062] In this situation, when calculating the similarity metric, an upper limit is set on the consecutive disappearance time of the target (e.g., within a preset time period before the current moment). Historical targets with a consecutive disappearance time less than this upper limit can be used for matching, but a penalty is added based on the consecutive disappearance time. This is done to ensure that historical targets that have disappeared for a period of time can be matched, but their validity in matching is weakened, that is, their competitiveness in matching the target being tested with other historical targets from more recent times is weakened.

[0063] For shape difference measurement, this embodiment considers the shape differences between targets when calculating the measurement parameters, but weakens the shape difference of small-scale target matching. At the same time, for target matching with low point cloud density, the calculation of its shape difference is relaxed, which can be considered as the validity of the observed target size being low when the point cloud density is low.

[0064] It should be noted that the smaller the value of the metric parameter, the higher the similarity between the two targets; conversely, the larger the value of the metric parameter, the lower the similarity. Furthermore, a historical target corresponding to a metric parameter is eligible for a match only if the value of the metric parameter is less than 1.

[0065] Step 104: Match the target to be tested against the historical targets based on the metric parameters of the target to be tested relative to each historical target.

[0066] In this embodiment, for each target to be tested, after calculating the metric parameter relative to each historical target, if the metric parameter is greater than or equal to 1, the metric parameter will be set to a maximum value (considered as abandoning the matching relationship). The metric parameters of all targets to be tested relative to historical targets constitute a similarity metric matrix, and then the Hungarian algorithm is used to obtain the matching relationship between the targets to be tested and historical targets.

[0067] The matching relationship between the target to be tested and the historical target can be that all the targets to be tested are successfully matched with the corresponding historical targets; or only some of the targets to be tested are matched with the corresponding historical targets; or none of the targets to be tested are successfully matched with the historical targets, and all the targets to be tested are appearing for the first time at the current moment.

[0068] For each target to be tested, once a target is successfully matched with a historical target, the historical target that matches the target is identified. The target to be tested inherits the ID of the historical target and is added to the trajectory of that historical target. In other words, the target identified at the current moment has appeared within a preset time period. This could mean it appeared continuously in every captured point cloud data, or it could have disappeared for a period and then reappeared in the target point cloud data. This target to be tested will inherit the trajectory and ID of the historical target it matches. That is, the historical target that appeared at the current moment is the target to be tested. The trajectory of the historical target is merged with the trajectory of the target to be tested at the current moment to form the inherited trajectory of the target to be tested. Here, the ID is a unique identifier for each target, which can be letters, numbers, or a combination of letters and numbers, etc., but this embodiment does not limit this.

[0069] If the match fails, a new trajectory and ID are created for the target. This means that the target did not appear in the previously preset time period, and it is the first time the target has been captured by the 3D sensing device. Of course, it is also possible that the target appeared before, but before the previously preset time period, and there is no corresponding historical target in the registration management.

[0070] In summary, the target matching method provided in this embodiment considers the impact of historical target disappearance time differences in the calculated measurement parameters when matching the target to be tested with historical targets. Historical targets that disappear within a preset time period can be matched, and a time difference measurement between the predicted target and the target to be tested based on the current moment is added for historical targets with longer disappearance times. On the other hand, the influence of shape differences is also considered, i.e., a shape difference measurement between the predicted target and the target to be tested based on the current moment is added, including cases where there is a large shape difference between the target to be tested and the predicted target. When calculating the shape difference measurement, the length and width of some targets to be tested are relaxed in a targeted manner according to differences in target detection validity, thereby improving the accuracy of target tracking. Furthermore, during matching, there are no direct matching restrictions based on the category of the target to be tested, avoiding target matching failures due to incorrect observation categories.

[0071] In one embodiment, determining the centroid distance metric of the target relative to the historical target may include:

[0072] The system acquires the position information of the first centroid of the predicted target, the position information of the second centroid of the target to be measured, and a preset centroid distance matching threshold; the position information includes the x-axis coordinate and the y-axis coordinate.

[0073] Based on the center-of-gravity distance matching threshold, the position information of the first center-of-gravity point, and the position information of the second center-of-gravity point, the center-of-gravity distance metric of the target to be tested relative to the historical target is determined.

[0074] Specifically, the predicted position information of the first centroid of the target refers to the position information predicted at the current moment based on the last moment the historical target appeared. In other words, the predicted position information of the first centroid of the target is predicted, not actually measured. In one implementation, Kalman filtering can be used to predict the current moment based on the last moment the historical target appeared in its trajectory, thereby obtaining the predicted position information of the first centroid of the target. The position information of the second centroid of the target to be measured can be calculated. The centroid distance matching threshold is a preset value.

[0075] In one alternative implementation, the centroid distance metric of the target to be tested relative to the historical target can be determined using the following formula, based on the centroid distance matching threshold, the first centroid position information, and the second centroid position information:

[0076] H ij重心 =[(x i ―x j ) 2 +(y i ―y j ) 2 ] 0.5 / s dlmt

[0077] Among them, H ij重心 x is the centroid distance metric for target i relative to historical target j, where i and j are both integers greater than 0; i Let x be the x-axis coordinate of the second centroid of the target i; j The x-axis coordinate of the first centroid of target j is predicted; y i Let y be the y-axis coordinate of the second centroid of the target i; j To predict the y-axis coordinate of the first centroid of target j; s dlmt This is the centroid distance matching threshold.

[0078] In one embodiment, determining a time difference measure of the target relative to a historical target may include:

[0079] The last moment the historical target appeared, the sampling period of the 3D sensing device, and the preset time difference factor were obtained.

[0080] Based on the current time, the last time the historical target appeared, the sampling period of the 3D sensing device, and the time difference factor, the time difference measure of the target to be measured relative to the historical target is determined.

[0081] The last moment when the historical target appears can be one of two possibilities: either it's the moment when the previous frame of point cloud data was captured by the 3D sensing device, which is adjacent to the current moment when the 3D sensing device sampled the data; or it could be a moment that is several sampling cycles apart from the current moment. In this embodiment, the time difference factor is a preset value that can be adjusted according to actual conditions.

[0082] In an optional implementation, the time difference measure between the target and the historical target is determined using the following formula, based on the current time, the last time the historical target appeared, the sampling period of the 3D sensing device, and the time difference factor:

[0083] H ij时间 =δ1·Relu(t) i ―t j ―T) / T

[0084] Among them, H ij时间 t is a measure of the time difference between target i and historical target j, where i and j are both integers greater than 0. i t represents the current time; j δ is the last moment when the historical target j appears; T is the sampling period; δ1 is the time difference factor, δ1≥0.

[0085] In one embodiment, determining a shape difference measure of the target to be measured relative to a historical target may include:

[0086] Obtain the first length, first width, and first point cloud density of the predicted target of the historical target at the last moment of its appearance; obtain the second length, second width, and second point cloud density of the target to be measured, as well as the preset first relaxation factor, preset second relaxation factor, preset shape difference factor, and standard point cloud density.

[0087] Based on the first length, first width, first point cloud density, second length, second width, second point cloud density, first relaxation term factor, second relaxation term factor, shape difference factor, and standard point cloud density, determine the shape difference measure of the target to be measured relative to the historical target.

[0088] The last moment of the historical target's appearance can also be understood as the moment when the historical target is closest to the current moment. The first relaxation term factor, the second relaxation term factor, the shape difference factor, and the standard point cloud density are all preset values.

[0089] Shape difference measurement is reflected in two aspects: length and width. Based on the first length, first point cloud density, second length, second point cloud density, first relaxation factor, second relaxation factor, shape difference factor, and standard point cloud density, the length difference measure of the target relative to the historical target can be determined using the following formula:

[0090]

[0091] Based on the first width, first point cloud density, second width, second point cloud density, first relaxation factor, second relaxation factor, shape difference factor, and standard point cloud density, the width difference measure of the target under test relative to the historical target can be determined using the following formula:

[0092]

[0093] Among them, H ij长度 H is used to measure the length difference between the target i and the historical target j. ij宽度 Let be the measure of the width difference between the target i and the historical target j, where i and j are both integers greater than 0; i Let l be the second length of the target i to be measured; j To predict the first length of target j; w i w represents the second width of the target i to be measured. j ρ is the first width of the predicted target j; stac δ is the standard point cloud density; δ4 is the shape difference factor, δ4≥0; γ is the first relaxation term factor, γ≥0; ε is the second relaxation term factor, ε≥0.

[0094] It is easy to understand, based on the above H ij重心 H ij时间 H ij长度 H ij宽度 The metric parameter H can be obtained. ij The full expression is not elaborated here.

[0095] In some scenarios, such as parking and waiting, the detection performance of multiple targets over long periods can be unstable, potentially leading to missed detections, false detections, and incorrect target category identification. Furthermore, in a stationary state, the predicted targets provided by Kalman filtering lack effective directionality. In such cases, simply matching based on target distance or improved intersection-over-union ratio (IoU) can easily result in mismatches with neighboring targets. If matching is performed based on target category, matching may fail due to changes in the target category. Therefore, when calculating similarity metrics, using centroid distance (the highest indicator of target detection validity), along with shape difference metrics (3D detection has different validity for shape detection at different target scales), and an additional penalty term for target disappearance time (i.e., temporal difference metric), can effectively optimize tracking stability under unstable target detection conditions. Furthermore, adding heading difference and point cloud density difference metrics can further optimize the stability of target tracking.

[0096] Therefore, in one embodiment, the metric parameter H ij It can also include a point cloud density difference metric H. ij密度 and / or heading difference measure H ij航向 .

[0097] In one implementation, determining a measure of the difference in point cloud density between the target to be measured and the historical target may include:

[0098] Obtain the preset point cloud density difference factor, the preset point cloud density difference threshold, the first point cloud density of the historical target at the last moment of its appearance, and the second point cloud density of the target to be measured.

[0099] Based on the point cloud density difference factor, point cloud density difference threshold, first point cloud density, and second point cloud density, the point cloud density difference measure of the target to be measured relative to the historical target is determined.

[0100] The first point cloud density is the point cloud density recorded and stored at the last moment of the historical target's appearance, which can be obtained from the point cloud data captured by the 3D sensing device at that last moment. The second point cloud density of the target to be measured can be obtained from the target point cloud data captured by the target to be measured. In this embodiment, the point cloud density difference factor and the point cloud density difference threshold are preset values.

[0101] Based on the point cloud density difference factor, point cloud density difference threshold, first point cloud density, and second point cloud density, the point cloud density difference measure of the target to be measured relative to the historical target can be determined using the following formula:

[0102] H ij 密度 =δ2·Relu(|ρ i ―ρ j |―ρdlmt ) / ρ dlmt

[0103] Among them, H ij密度 ρ is a measure of the point cloud density difference between the target i and the historical target j, where i and j are both integers greater than 0. dlmt δ is the point cloud density difference threshold, δ2 is the point cloud density difference factor, δ2≥0, ρ i For the second point cloud density, ρ j The first point is the cloud density.

[0104] In one implementation, determining the heading difference measure of the target to be measured relative to the historical target may include:

[0105] Acquire the preset heading difference factor, preset standard angle change rate, preset angle difference threshold, heading angle of the target to be measured, heading angle of the predicted target of the historical target, and the last time the historical target appeared;

[0106] Based on the current time, heading difference factor, standard angle change rate, angle difference threshold, heading angle of the target to be measured, heading angle of the predicted target of the historical target, and the last time the historical target appeared, determine the heading difference measure of the target to be measured relative to the historical target.

[0107] In this embodiment, the heading difference factor, standard angle change rate, and angle difference threshold are all preset values. The heading angle of the target to be measured can be obtained from the target point cloud data captured at the current moment. The heading angle of the predicted historical target is the predicted data.

[0108] Based on the current time, heading difference factor, standard angle change rate, angle difference threshold, heading angle of the target under test, heading angle of the predicted target of the historical target, and the last time the historical target appeared, the heading difference measure of the target under test relative to the historical target is determined by the following formula:

[0109] H ij航向 =δ3·Relu[min(|θ i -θ j |,2π―|θ i -θ j |)-d θ ·(t i -t j )] / θ dlmt

[0110] Among them, H ij航向 θ is the measure of the heading difference between the target i and the historical target j, where i and j are both integers greater than 0. i Let θ be the heading angle of the target to be measured, 0 ≤ θ i<2π, θ j Let θ be the heading angle of the predicted target for this historical target, 0 ≤ θ j <2π, δ3 is the heading difference factor, δ3≥0, t i For the current time, t j For the last moment when the historical goal j appears, θ dlmt d is the threshold for angle difference. θ This represents the standard angle change rate.

[0111] In summary, the target matching method provided in this specification considers the impact of historical target disappearance time differences when matching a target to be tested with historical targets. Historical targets that disappear within a preset time period can be matched. The method also incorporates a time difference metric between the predicted target and the target at the current moment for historical targets with longer disappearance times. Furthermore, it considers the impact of shape differences, including a shape difference metric between the predicted target and the target at the current moment for historical targets, covering cases with significant shape differences between the two. During the shape difference metric calculation, relaxation is applied to small-scale targets and targets with low point cloud density based on differences in target detection validity, thereby improving the accuracy of target tracking.

[0112] Based on the same inventive concept, combined with Figure 2 As shown, embodiments of the present invention also provide a target matching device, comprising:

[0113] The acquisition module is used to acquire historical targets and predicted targets based on the current time. Historical targets are point cloud targets detected and registered for management within a preset time period before the current time.

[0114] The determination module is used to determine the target to be measured based on the target point cloud data collected by the 3D sensing device at the current moment;

[0115] The calculation module is used to determine, for each predicted target of a historical target, a metric parameter to characterize the similarity between the target to be tested and the predicted target of the historical target, based on the target to be tested and the predicted target of the historical target. The metric parameter includes a centroid distance metric to characterize the similarity of the centroid positions, a time difference metric to characterize the continuous disappearance time of the historical target, and a shape difference metric to characterize the length and width similarity between the target to be tested and the historical target, wherein the shape difference metric is determined based on the point cloud density of the target to be tested and the historical target.

[0116] The matching module is used to match the target to be tested against historical targets based on the metric parameters of the target to be tested relative to each historical target.

[0117] In one alternative implementation, the computing module is further configured to:

[0118] The system acquires the position information of the first centroid of the predicted target, the position information of the second centroid of the target to be measured, and a preset centroid distance matching threshold; the position information includes the x-axis coordinate and the y-axis coordinate.

[0119] Based on the center-of-gravity distance matching threshold, the position information of the first center-of-gravity point, and the position information of the second center-of-gravity point, the center-of-gravity distance metric of the target to be tested relative to the historical target is determined.

[0120] In one alternative implementation, the computing module is further configured to:

[0121] The last moment the historical target appeared, the sampling period of the 3D sensing device, and the preset time difference factor were obtained.

[0122] Based on the current time, the last time the historical target appeared, the sampling period of the 3D sensing device, and the time difference factor, the time difference measure of the target to be measured relative to the historical target is determined.

[0123] In one alternative implementation, the computing module is further configured to:

[0124] Obtain the first length, first width, and first point cloud density of the predicted target of the historical target at the last moment of its appearance; obtain the second length, second width, and second point cloud density of the target to be measured, as well as the preset first relaxation factor, preset second relaxation factor, preset shape difference factor, and standard point cloud density.

[0125] Based on the first length, first width, first point cloud density, second length, second width, second point cloud density, first relaxation term factor, second relaxation term factor, shape difference factor, and standard point cloud density, determine the shape difference measure of the target to be measured relative to the historical target.

[0126] Optionally, the metrics may also include point cloud density difference metrics and / or heading difference metrics.

[0127] In one alternative implementation, the computing module is further configured to:

[0128] Obtain the preset point cloud density difference factor, the preset point cloud density difference threshold, the first point cloud density of the historical target at the last moment of its appearance, and the second point cloud density of the target to be measured.

[0129] Based on the point cloud density difference factor, point cloud density difference threshold, first point cloud density, and second point cloud density, the point cloud density difference measure of the target to be measured relative to the historical target is determined.

[0130] In one alternative implementation, the computing module is further configured to:

[0131] Acquire the preset heading difference factor, preset standard angle change rate, preset angle difference threshold, heading angle of the target to be measured, heading angle of the predicted target of the historical target, and the last time the historical target appeared;

[0132] Based on the current time, heading difference factor, standard angle change rate, angle difference threshold, heading angle of the target to be measured, heading angle of the predicted target of the historical target, and the last time the historical target appeared, determine the heading difference measure of the target to be measured relative to the historical target.

[0133] In an alternative implementation, the matching module is further configured to:

[0134] If a match is successful, the historical target that matches the target to be tested is identified. The target to be tested inherits the ID of the historical target and is added to the trajectory of the historical target.

[0135] If the match fails, a new trajectory and ID will be created for the target to be tested.

[0136] In one alternative implementation, the computing module is further configured to:

[0137] Based on the target to be tested and the predicted target of that historical target, the following formula is used to determine the metric parameter used to characterize the similarity between the target to be tested and the historical target:

[0138] H ij =[(x i ―x j ) 2 +(y i ―y j ) 2 ] 0.5 / s dlmt +

[0139] δ1·Relu(t i ―t j ―T) / T+

[0140]

[0141] Among them, H ij x is the measurement parameter of target i relative to historical target j; i Let x be the x-axis coordinate of the second centroid of the target i; j The x-axis coordinate of the first centroid of target j is predicted; y i Let y be the y-axis coordinate of the second centroid of the target i; j To predict the y-axis coordinate of the first centroid of target j; l i Let l be the length of the target i to be measured; j To predict the length of target j; wi w is the width of the target i to be measured; j To predict the width of target j; t i t represents the current time; j The last moment that historical target j appears; s dlmt The centroid distance matching threshold; ρ stac δ is the standard point cloud density; T is the sampling period; δ1 is the time difference factor, δ1≥0; δ4 is the shape difference factor, δ4≥0; γ is the first relaxation term factor, γ≥0; ε is the second relaxation term factor, ε≥0.

[0142] In summary, the target matching device provided in this specification, when matching a target to be tested with historical targets, calculates measurement parameters that take into account the impact of differences in the disappearance time of historical targets. Historical targets that disappear within a preset time period can be matched, and a time difference measurement between the predicted target and the target to be tested based on the current moment is added for historical targets with longer disappearance times. On the other hand, the influence of shape differences is also considered, i.e., a shape difference measurement between the predicted target and the target to be tested based on the current moment is added, including cases where there is a large shape difference between the target to be tested and the predicted target. During the shape difference measurement calculation, relaxation is applied to small-scale targets and targets with low point cloud density levels based on differences in target detection validity, thereby improving the accuracy of target tracking.

[0143] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the target matching device described above can be referred to the corresponding process in the aforementioned method, and will not be elaborated further here.

[0144] Based on the above, this embodiment provides a readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the target matching method of any of the foregoing embodiments.

[0145] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the readable storage medium described above can be referred to the corresponding process in the aforementioned method, and will not be elaborated further here.

[0146] The above are merely various embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A target matching method, characterized in that, include: Obtain historical targets and predicted targets based on those historical targets at the current moment; The historical targets are point cloud targets detected and registered for management within a preset time period before the current moment; The target to be measured is determined based on the target point cloud data collected by the 3D sensing device at the current moment; For each predicted target of a historical target, based on the target to be tested and the predicted target of the historical target, a metric parameter is determined to characterize the similarity between the target to be tested and the historical target. The metric parameter includes a centroid distance metric to characterize the similarity of centroid positions, a temporal difference metric to characterize the continuous disappearance time of the historical target, and a shape difference metric to characterize the length and width similarity between the target to be tested and the historical target, wherein the shape difference metric is determined based on the point cloud density of the target to be tested and the historical target. Based on the measurement parameters of the target to be tested relative to each historical target, the target to be tested is matched among the historical targets; Determining the time difference metric of the target to be measured relative to the historical target includes: The last moment the historical target appeared, the sampling period of the three-dimensional sensing device, and the preset time difference factor were obtained. Based on the current time, the last time the historical target appeared, the sampling period of the three-dimensional sensing device, and the time difference factor, the time difference measure of the target to be measured relative to the historical target is determined. Determining the shape difference measure of the target object relative to the historical target includes: The first length, first width, and first point cloud density of the predicted target of the historical target at the last moment of its appearance are obtained. The second length, second width, and second point cloud density of the target to be measured, the preset first relaxation factor, the preset second relaxation factor, the preset shape difference factor, and the preset standard point cloud density are obtained. The shape difference measure of the target under test relative to the historical target is determined based on the first length, the first width, the first point cloud density, the second length, the second width, the second point cloud density, the first relaxation term factor, the second relaxation term factor, the shape difference factor, and the standard point cloud density.

2. The method according to claim 1, characterized in that, Determining the centroid distance metric of the target to be measured relative to the historical target includes: Obtain the first centroid position information of the predicted target of the historical target, the second centroid position information of the target to be tested, and the preset centroid distance matching threshold; Based on the center-of-gravity distance matching threshold, the first center-of-gravity position information, and the second center-of-gravity position information, the center-of-gravity distance metric of the target to be tested relative to the historical target is determined.

3. The method according to claim 1 or 2, characterized in that, The metric parameters also include point cloud density difference metric and / or heading difference metric.

4. The method according to claim 3, characterized in that, Determining the point cloud density difference metric of the target object relative to the historical target includes: Obtain a preset point cloud density difference factor, a preset point cloud density difference threshold, the first point cloud density of the historical target at the last moment of its appearance, and the second point cloud density of the target to be tested; The point cloud density difference measure of the target under test relative to the historical target is determined based on the point cloud density difference factor, the point cloud density difference threshold, the first point cloud density, and the second point cloud density.

5. The method according to claim 3, characterized in that, Determining the heading difference metric of the target under test relative to the historical target includes: The system acquires a preset heading difference factor, a preset standard angle change rate, a preset angle difference threshold, the heading angle of the target to be measured, the heading angle of the predicted target of the historical target, and the last moment when the historical target appeared. The heading difference measure of the target under test relative to the historical target is determined based on the current time, the heading difference factor, the standard angle change rate, the angle difference threshold, the heading angle of the target under test, the heading angle of the predicted target of the historical target, and the last time the historical target appeared.

6. The method according to claim 1, characterized in that, The method further includes: If a match is successful, a historical target that matches the target to be tested is identified. The target to be tested inherits the ID of the historical target and is added to the trajectory of the historical target. If the matching fails, a new trajectory and ID are created for the target to be tested.

7. The method according to claim 1, characterized in that, The step of determining a metric parameter to characterize the similarity between the target to be tested and the predicted target of the historical target, based on the target to be tested and the predicted target of the historical target, includes: Based on the target to be tested and the predicted target of the historical target, the following formula is used to determine the metric parameter characterizing the similarity between the target to be tested and the historical target: in, Let i be the measurement parameter of the target i relative to the historical target j, where i and j are both integers greater than 0; For the target to be tested The second center of gravity Axis coordinates; For predicting the target The first point of focus Axis coordinates; For the target to be tested The second center of gravity Axis coordinates; For predicting the target The first point of focus Axis coordinates; For the target to be tested Length; For predicting the target Length; For the target to be tested The width; For predicting the target The width; The current moment; For this historical goal The last moment it appeared; The centroid distance matching threshold; Standard point cloud density; The sampling period; As a time difference factor, ; For shape difference factor, ; The first relaxation term factor, ; The second relaxation term factor, .

8. A target matching device, characterized in that, The target matching method according to any one of claims 1-7; the apparatus comprises: The acquisition module is used to acquire historical targets and predicted targets of the historical targets based on the current time; the historical targets are point cloud targets detected and registered for management within a preset time period before the current time. The determination module is used to determine the target to be measured based on the target point cloud data collected by the 3D sensing device at the current moment; The calculation module is used to determine, for each predicted target of a historical target, a metric parameter to characterize the similarity between the target to be tested and the predicted target of the historical target, based on the target to be tested and the predicted target of the historical target. The metric parameter includes a centroid distance metric to characterize the similarity of the centroid positions, a time difference metric to characterize the continuous disappearance time of the historical target, and a shape difference metric to characterize the length and width similarity between the target to be tested and the historical target, wherein the shape difference metric is determined based on the point cloud density of the target to be tested and the historical target. The matching module is used to match the target to be tested with each historical target based on the measurement parameters of the target to be tested relative to each historical target.

Images