Target tracking method and device based on adaptive correlation gate, and storage medium

Through adaptive correlation wavegate and multi-dimensional weighted probability calculation, the problem of insufficient correlation accuracy of the target tracking algorithm in the prior art in the multi-target environment is solved, which improves the stability and accuracy of target tracking and reduces the calculation amount.

CN119025798APending Publication Date: 2024-11-26COWA TECHNOLOGY CO LTD +1
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411141029.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The existing target tracking algorithms have insufficient correlation accuracy in multi-target environments, resulting in false target generation, especially in target-intensive scenarios, resulting in poor tracking performance, and fixed wave gate settings lead to an increase in the computational volume, affecting stability.

Method used

The target tracking method based on adaptive correlation wave gate is adopted to dynamically adjust the size and shape of the correlation wave gate according to the tracking target type and motion state, and calculate the correlation probability through multi-dimensional information weighting to improve correlation accuracy and stability.

Benefits of technology

It improves the probability that the true measurement target falls on the correlation wave gate, reduces the candidate target data, reduces the amount of calculation, enhances the tracking stability and accuracy, and reduces the target correlation error rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025798A_ABST
    Figure CN119025798A_ABST
Patent Text Reader

Abstract

The invention discloses a target tracking method and device based on an adaptive correlation gate, and a storage medium. The method comprises the steps of determining a tracking target motion state and a filtering model according to a tracking target type; determining an associated gate according to the type of the tracking target and the motion state of the tracking target; calculating the association probability between the tracking target and the measurement target in the association gate, and setting the association probability to 0 for the measurement target which does not fall into the association gate to obtain an association probability matrix; calculating a distribution relation between the tracking target and the measurement target based on the association probability matrix between the tracking target and the measurement target; re-calculating the Euclidean distance between the tracking target and the measurement target which are not distributed, generating a new association probability matrix, and calculating a new distribution relation; and updating the tracking target motion state and the filtering model of the tracking target according to the latest distribution relation. According to the invention, the stability and accuracy of target tracking can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and relates to target association and target tracking methods, and specifically to a target tracking method, device and storage medium based on an adaptive association gate. Background Art

[0002] With the development of autonomous driving technology, driving safety is a key link. In order to perceive the driving environment, autonomous driving vehicles are usually equipped with multiple sensors (cameras, lidar, millimeter-wave radar, etc.). By processing and analyzing sensor data, they can obtain information such as the surrounding environment, targets, obstacles, etc., to provide support for autonomous driving planning and control.

[0003] Autonomous driving perception technology requires tracking and trajectory prediction of detected targets. Therefore, multi-target tracking technology is also one of the hot topics in current autonomous driving perception technology research. Data association between multiple targets is an extremely important link in multi-target tracking. The accuracy of data association will directly affect the performance of tracking and predicting the target's true trajectory. If the association error rate is high, many false targets will be generated; only by improving the accuracy of association can the actual number of targets in the real scene be restored as much as possible.

[0004] At present, the commonly used association algorithms include the Hungarian algorithm, the nearest neighbor (NN) algorithm, the probability density association (PDA) algorithm, the joint probability density association (JPDA) algorithm, etc.

[0005] The nearest neighbor method first sets an association gate, and the target obtained by the preliminary screening of the association gate becomes a candidate target to limit the number of related judgments. The association gate is a subspace in the tracking space, and its center is located at the predicted position of the tracked target. The size of the association gate should be designed to ensure that the correct measurement target falls into the association gate with a certain probability. If there is only one measurement falling into the association gate, the measurement value can be directly used to update the target trajectory; but if there is more than one measurement falling into the association gate of the tracked target, the measurement target with the smallest statistical distance should be taken as the updated trajectory. The advantage of the nearest neighbor method is that the calculation is simple, and the disadvantage is that the candidate target closest to the predicted position of the target in a multi-target environment is not necessarily the real target of the target. Therefore, it is suitable for target tracking in a sparse echo environment.

[0006] The Hungarian algorithm is a method for solving the minimum cost allocation problem. By calculating the cost matrix of historical tracking targets and measurement targets, the Hungarian algorithm is used to solve the global minimum cost and thus realize the association between historical tracking targets and measurement targets. However, the process of solving the cost matrix does not consider the probability of association between targets, so the cost matrix is ​​obtained by an indiscriminate calculation, such as calculating the IOU or Euclidean distance between targets. In a scene with dense targets, the Hungarian matching algorithm usually selects the target that is closest to the target. This will cause the target ID to jump frequently, which is not conducive to tracking stability.

[0007] The probabilistic data association (PDA) algorithm considers all candidate targets that fall into the correlation gate, and calculates the probability of each target coming from the target according to different correlation situations. Then, these probability values ​​are used to weight the different targets in the correlation gate. The weighted sum of each candidate target is used as an equivalent target, and the equivalent target is used to update the state of the target. The probabilistic data association algorithm is a suboptimal filtering method that only decomposes the latest measurement and is mainly used to solve the problem of single radar single target tracking in a cluttered environment.

[0008] Since the PDA algorithm cannot accurately consider the impact of common targets in the intersection area of ​​multiple target association gates on track updating, the tracking performance of PDA is not ideal when the targets are densely packed.

[0009] The JPDA algorithm uses the position point of the current measurement target that falls within the association gate to calculate the association probability between the position point and the corresponding trajectory, and uses the association probability to calculate the weighted sum of the position point of the current measurement target to correct the trajectory. The weight is the probability that the measurement target position point in the tracking comes from the tracking target. In actual calculation, the weight is obtained by finding the combination set of all possible measurement target position points and tracking target trajectories, and calculating the probability of the association set. The difference between it and PDA is that the calculation of probability interconnection probability is different.

[0010] The existing association algorithm in target tracking uses a fixed gate to calculate the association probability between targets. If the gate is set too small, the real measurement target is easily lost. If the gate is set too large, the number of candidate measurement targets increases, which increases the algorithm's computational complexity. Summary of the invention

[0011] In order to solve the above technical problems, the present invention proposes a target tracking method, device and storage medium based on adaptive correlation gate.

[0012] In order to achieve the above object, the technical solution of the present invention is as follows:

[0013] In a first aspect, the present invention discloses a target tracking method based on an adaptive correlation gate, comprising:

[0014] Step S1: Determine the motion state and filtering model of the tracking target according to the tracking target type;

[0015] Step S2: determining the associated wave gate according to the tracking target type and the tracking target motion state;

[0016] Step S3: Calculate the association probability between the tracking target and the measurement target within the association wave gate. For the measurement target that does not fall into the association wave gate, the association probability is set to 0, and the association probability matrix is ​​obtained;

[0017] Step S4: calculating the distribution relationship between the tracking target and the measurement target based on the association probability matrix between the tracking target and the measurement target;

[0018] Step S5: for the unassigned tracking targets and measurement targets, recalculate the Euclidean distances between them, generate a new association probability matrix, and calculate a new assignment relationship;

[0019] Step S6: According to the latest allocation relationship, the tracking target motion state and the filter model of the tracking target are updated.

[0020] Based on the above technical solution, the following improvements can be made:

[0021] As a preferred solution, step S2 includes:

[0022] Step S2.1: determining the correlation gate model according to the tracking target type and the tracking target motion state;

[0023] If the tracking target type is a stationary target, or the tracking target type is a moving target and the moving state of the tracking target is stationary, a circular correlation wave gate of a fixed size is selected to obtain a correlation wave gate model;

[0024] If the tracking target type is a moving target and the tracking target motion state is moving, then the elliptical correlation wave gate that conforms to the current tracking target motion state is calculated to obtain a correlation wave gate model;

[0025] Step S2.2: Bring the measurement target set of the current frame into the correlation gate model to determine whether it falls into the correlation gate.

[0026] As a preferred solution, the following steps are performed to calculate the elliptical correlation wave gate that meets the motion state of the current tracking target, specifically including:

[0027] Step A: Based on the motion state of the tracking target, the orientation angle of the motion speed of the tracking target is calculated to obtain an elliptical correlation wave gate with an adaptive rotation direction, and the major axis direction of the elliptical correlation wave gate is consistent with the velocity direction of the tracking target;

[0028] Step B: Calculate the major and minor axis radii of the elliptical correlation wave gate according to the motion state of the tracking target.

[0029] As a preferred solution, in step S3, the association probability is a weighted association probability based on multi-dimensional information.

[0030] In a second aspect, the present invention discloses a target tracking device based on an adaptive correlation gate, comprising:

[0031] A tracking target determination module is used to determine the tracking target motion state and filtering model according to the tracking target type;

[0032] A correlation wave gate determination module is used to determine the correlation wave gate according to the tracking target type and the tracking target motion state;

[0033] The association probability calculation module is used to calculate the association probability between the tracking target and the measurement target within the association wave gate. For the measurement target that does not fall into the association wave gate, the association probability is set to 0 to obtain the association probability matrix;

[0034] A distribution relationship preliminary calculation module, used for calculating the distribution relationship between the tracking target and the measurement target based on the association probability matrix between the tracking target and the measurement target;

[0035] The allocation relationship secondary calculation module is used to recalculate the Euclidean distance between unallocated tracking targets and measurement targets, generate a new association probability matrix, and calculate a new allocation relationship;

[0036] The updating module is used to update the tracking target motion state and the filtering model of the tracking target according to the latest allocation relationship.

[0037] As a preferred solution, the associated wave gate determination module includes:

[0038] A correlation wave gate model determination unit, used to determine the correlation wave gate model according to the tracking target type and the tracking target motion state;

[0039] If the tracking target type is a stationary target, or the tracking target type is a moving target and the moving state of the tracking target is stationary, a circular correlation wave gate of a fixed size is selected to obtain a correlation wave gate model;

[0040] If the tracking target type is a moving target and the tracking target motion state is moving, then the elliptical correlation wave gate that conforms to the current tracking target motion state is calculated to obtain a correlation wave gate model;

[0041] The judgment unit is used to bring the measurement target set of the current frame into the correlation wave gate model to judge whether it falls into the correlation wave gate.

[0042] As a preferred solution, the correlation wave gate determination module further includes: an elliptical correlation wave gate calculation unit, which is used to calculate an elliptical correlation wave gate that conforms to the current tracking target motion state, specifically including:

[0043] A rotation direction calculation unit is used to calculate the direction angle of the tracking target's motion speed based on the tracking target's motion state, and obtain an elliptical correlation wave gate with an adaptive rotation direction, wherein the major axis direction of the elliptical correlation wave gate is consistent with the speed direction of the tracking target;

[0044] The radius calculation unit is used to calculate the major and minor axis radii of the elliptical correlation wave gate according to the motion state of the tracking target.

[0045] As a preferred solution, in the association probability calculation module, the association probability is a weighted association probability based on multi-dimensional information.

[0046] In a third aspect, the present invention discloses a storage medium storing one or more computer-readable programs, wherein the one or more programs include instructions suitable for being loaded by a memory and executing the above-mentioned target tracking method based on adaptive correlation gate.

[0047] The present invention discloses a target tracking method, device and storage medium based on adaptive correlation wave gate, which adaptively updates the correlation wave gate based on the motion state of the tracked target, improves the probability that the real measured target falls into the correlation wave gate, reduces the candidate target data in the correlation wave gate, improves the correlation accuracy, reduces the amount of calculation, improves the tracking stability, and improves the algorithm processing rate. At the same time, a method for calculating the weighted correlation probability using multi-dimensional information is used to improve the target correlation accuracy, thereby improving the stability and accuracy of target tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 Schematic diagram of target association gate provided by prior art.

[0050] Figure 2 JPDA algorithm flow chart provided for prior art.

[0051] Figure 3 A flow chart of a target tracking method provided by an embodiment of the present invention.

[0052] Figure 4 A flowchart of determining an associated wave gate provided by an embodiment of the present invention.

[0053] Figure 5 A schematic diagram of an associated wave gate for a stationary target provided by an embodiment of the present invention.

[0054] Figure 6 Schematic diagram of the constant direction elliptical model associated wave gate provided by the prior art.

[0055] Figure 7 A schematic diagram of an adaptive elliptical correlation wave gate provided in an embodiment of the present invention.

[0056] Figure 8 A schematic diagram of IOU calculation provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] Using ordinal numbers “first,” “second,” “third,” etc. to describe common objects merely indicates that different instances of similar objects are involved and is not intended to imply that the objects so described must have a given order in time, space, order, or in any other manner.

[0060] In addition, the expression of “comprising” an element is an “open” expression, which merely means that corresponding components or steps exist, and should not be interpreted as excluding additional components or steps.

[0061] When the existing JPDA-based methods associate targets, they do not consider the actual motion state and target category. For example, the thresholds of the gate ranges for moving targets and stationary targets are set the same. When the threshold is set too large, the gate becomes larger. For stationary targets, the distance range of the measured target offset from the actual position is smaller, which will increase the amount of calculation when calculating the probability of association between targets.

[0062] like Figure 1 As shown in the figure, it is assumed that target A is a static target for historical tracking, and the others are measurement targets at the current moment. When the gate is set too large, such as Figure 1 As shown in (a), there are many candidate targets for the associated gate. If we consider that the target's historical tracking state is static, we can reduce the target's gate area (such as Figure 1 (b) shows that the number of candidate targets is reduced, thereby reducing the amount of computation.

[0063] The existing technical solution takes the JPDA algorithm process as an example. The general algorithm process is as follows Figure 2 As shown, the process steps are described as follows:

[0064] Step 1: Based on the state at the historical time t-1, predict the motion state at the current time t;

[0065] Step 2: Establish an association gate near the predicted position of each tracked target;

[0066] Step 3: According to the association gate, calculate the effective matrix Ω of the measurement target falling into the association gate;

[0067] Step 4: Generate feasible joint events from Ω and calculate the probability of feasible joint events;

[0068] Step 5: Calculate the association probability;

[0069] Step 6: According to the association probability, select the measurement target and update the status information of the tracking target.

[0070] In order to achieve the purpose of the present invention, some embodiments of the target tracking method, device and storage medium based on adaptive correlation gate are as follows: Figure 3 As shown, the target tracking method includes:

[0071] Step S1: Determine the motion state and filtering model of the tracking target according to the tracking target type;

[0072] Step S2: determining the associated wave gate according to the tracking target type and the tracking target motion state;

[0073] Step S3: Calculate the association probability between the tracking target and the measurement target within the association wave gate. For the measurement target that does not fall into the association wave gate, the association probability is set to 0, and the association probability matrix is ​​obtained;

[0074] Step S4: Based on the association probability matrix between the tracking target and the measurement target, the Hungarian algorithm is used to calculate the distribution relationship between the tracking target and the measurement target;

[0075] Step S5: for the unassigned tracking targets and measurement targets, recalculate the Euclidean distances between them, generate a new association probability matrix, and use the Hungarian algorithm to calculate a new assignment relationship;

[0076] Step S6: According to the latest allocation relationship, the tracking target motion state and the filter model of the tracking target are updated.

[0077] Each step of the present invention is described in detail below.

[0078] Step S1, determining the motion state and filtering model of the tracking target according to the tracking target type.

[0079] Currently, the target motion state models for tracking include constant velocity model (Constant Velocity, CV), constant acceleration model (Constant Acceleration, CA), constant turn rate and velocity model (Constant Turn Rate And Velocity, CTRV), constant turn rate and acceleration model (Constant Turn Rate And Velocity, CTRA), etc.

[0080] The filter model is divided into a linear Kalman filter model and a nonlinear Kalman filter model, wherein the nonlinear Kalman filter model includes an extended Kalman filter model (EKF) and an unscented Kalman filter model (UKF).

[0081] First, the motion state of the tracking target is determined according to the tracking target type.

[0082] For tracking targets such as pedestrians or other stationary targets, the CV model is used to simplify the motion process. The CV motion equation is defined as follows:

[0083]

[0084] The CV model assumes that the speed is constant within the time Δt, so the speed from time t-1 to time t is is constant; for the position x at time t k ,y k It is related to speed and reflects the change in displacement within the time Δt.

[0085] For non-motor vehicles, motor vehicles or other moving targets, the CTRV model is used to describe its nonlinear motion state, and its state equation is defined as follows:

[0086] Define the vehicle's yaw angle as θ. When the angular velocity change rate ω≠0, the motion model equation is defined as follows:

[0087]

[0088] When ω=0, the motion model equation is defined as follows:

[0089]

[0090] Where: x t ,y t , v t ,θ t ,ω t They represent the x-position, y-position, speed, yaw angle and angular velocity change rate at time t respectively.

[0091] Then, the filtering model is determined.

[0092] For both linear and nonlinear filtering models, they can be described as follows.

[0093] Assume that there is a target θ i , i = 1, 2, ..., m, according to the initial state filtering to the time t-1, the motion state at this time is recorded as X θi (t-1), the state covariance matrix is ​​recorded as P θi (t-1), the target transfer matrix at time t is recorded as F θi (t).

[0094] Predict the target motion state at time t

[0095] X θi (t|t-1)=F θi (t)X θi (t-1) Formula (4)

[0096] Predict the target motion state covariance at time t:

[0097] P θi (t|t-1)=F θi (t)(t-1)F θi (t) T +Q θi (t) Formula (5)

[0098] Where Q θi (t) is the predicted noise covariance at time t.

[0099] The predicted position at time t is obtained as:

[0100] Z θi (t|t-1)=H θi (t)X θi (t-1) Formula (6)

[0101] Z θi (t|t-1) is the target θ i At the predicted position at time t, H θi is the target θ i The observation matrix.

[0102] Update the target motion state at time t, Kalman gain K θi (t) calculation, where R θi is the target θ i The process noise matrix is:

[0103]

[0104] Obtain the target filter position X at time tθi (t) and the state covariance matrix P θi (t) is:

[0105] X θi (t) = X θi (t|t-1)+K θi (t|t-1)*(Z θi (t|t-1)-H θi (t)*X θi (t|t-1))

[0106] Formula (8)

[0107] P θi (t)=(IK θi (t)H θi (t))*P θi (t|t-1) Formula (9)

[0108] Step S2 is to adaptively calculate the correlation gate area. When the tracking target is associated with the measurement target, if a fixed-size correlation gate is used, when the motion data changes, the size of the correlation gate cannot be updated with the update of the motion data, which will cause the target association to fail and the target to be lost. Therefore, in order to improve the accuracy of target association, the present invention proposes a method for adaptively updating the correlation gate based on the motion state of the tracking target, thereby increasing the probability that the real measurement target falls within the correlation gate, improving the association accuracy, and thus improving the tracking stability.

[0109] like Figure 4 As shown, step S2 includes:

[0110] Step S2.1: determining the correlation gate model according to the tracking target type and the tracking target motion state;

[0111] If the tracking target type is a stationary target, or the tracking target type is a moving target and the moving state of the tracking target is stationary, a circular correlation wave gate of a fixed size is selected to obtain a correlation wave gate model;

[0112] If the tracking target type is a moving target and the tracking target motion state is moving, then the elliptical correlation wave gate that conforms to the current tracking target motion state is calculated to obtain a correlation wave gate model;

[0113] Step S2.2: Bring the measurement target set of the current frame into the correlation gate model to determine whether it falls into the correlation gate.

[0114] Target tracking involves two types of tracking: static targets and moving targets. For different motion states, the gate design used for association is also different. For static targets, the distribution direction of the detection error is uncertain, so a circular correlation gate is designed to associate it; for moving targets, the distribution of the detection error mainly exists in the direction of motion of the main axis, and the detection error for another direction of motion perpendicular to the main axis is small, so it is more suitable for an elliptical type of correlation gate.

[0115] The circular correlation gate design for stationary targets is as follows:

[0116] When the target is stationary, the speed direction is invalid. It is necessary to build an elliptical model with the help of the direction angle given by the detection end. At this time, it is necessary to consider the target category information. If the target is a stationary obstacle such as a roadblock or tree trunk, or a stationary pedestrian, when designing the elliptical model, considering the inaccurate detection direction, the elliptical model can be degraded into a circular model to improve the accuracy of target association, such as Figure 5 shown.

[0117] (x i -x0) 2 +(y i -y0) 2 ≤R 2 Formula (10)

[0118] For a moving target, and the tracking target motion state is moving, the following steps are performed to calculate an elliptical correlation wave gate that conforms to the current tracking target motion state, specifically including:

[0119] Step A: Based on the motion state of the tracking target, the orientation angle of the motion speed of the tracking target is calculated to obtain an elliptical correlation wave gate with an adaptive rotation direction, and the major axis direction of the elliptical correlation wave gate is consistent with the velocity direction of the tracking target;

[0120] Step B: Calculate the major and minor axis radii of the elliptical correlation wave gate according to the motion state of the tracking target.

[0121] Specifically, according to Z θi (t|t-1) is the center of the circle, and the design target θ i Elliptical correlation gate at time t.

[0122] Typically, the motion model of the considered target is defined as follows:

[0123]

[0124] Where: P x (t), P y (t) are the target θ i The coordinate position in the x-axis and y-axis directions at time t;

[0125] Δt is the time difference between time t and time t-1;

[0126] V x|t-1 、V y|t-1 are the target θ i The speed in the x-axis and y-axis directions at time t;

[0127] a x|t-1 、a y|t-1 are the target θ i The acceleration in the x-axis and y-axis directions at time t.

[0128] The velocity heading angle is as follows:

[0129] theta t =arctan(V y|t-1 / V x|t-1 ) Formula (12)

[0130] The ellipse model is defined as follows:

[0131]

[0132] Among them: x0, y0 are the centers of the ellipse, that is, the predicted positions of the target;

[0133] x i ,y i is the position of the measurement target to be calculated;

[0134] Ra and Rb are the major and minor axis radii of the elliptical model respectively.

[0135] Usually, the designed ellipse model is as follows Figure 6 As shown in the figure (the center of the red arrow is not in the associated wave gate), it does not conform to the actual operating state of the target. It is necessary to stipulate that the direction of the major axis of the ellipse is always consistent with the target speed direction. In order to achieve the purpose of consistency between the major axis of the ellipse and the speed direction, the standard elliptical wave gate model needs to be rotated with the help of a rotation matrix.

[0136] The rotation matrix is ​​shown in formula 14:

[0137]

[0138] Where: θ is the orientation angle of the target motion state.

[0139] Rotate the ellipse with the center (x0, y0) and the major and minor axis radii Ra and Rb respectively along mat The formula 15 is obtained by rotation, where (x', y') is the position of the center of the ellipse after rotation:

[0140]

[0141]

[0142] like Figure 7 As shown in the figure, adding a rotating ellipse model will improve the association effect of the target turning state (the center point of the red arrow is located in the association gate).

[0143] The present invention adaptively updates the direction of the elliptical correlation gate based on the motion state of the tracking target, and its application scenarios include but are not limited to campus scenarios, vehicle-mounted fields, traffic radar fields and other scenarios, and the above scenarios should all fall within the scope of protection.

[0144] When the target is actually moving, it may be in a state of stillness, acceleration / deceleration, uniform speed (low speed, high speed), etc. Therefore, when predicting the target position, different speeds will lead to different changes in the associated threshold. For example, when driving in a straight line, the main displacement direction is the front direction, and the displacement change in the body direction is small; but when the target is in the process of changing lanes or turning, the displacement difference in the body direction is higher than that in the front direction, so the settings of Ra and Rb will change. In the first case, the value of the major axis Ra is greater than the value of the minor axis Rb. For the second case, on the contrary, the value of Rb is greater than the value of Ra.

[0145] Considering the measurement error and prediction error of the sensor, the designed correlation gate threshold should be greater than the actual estimated displacement difference. The setting formulas of Ra and Rb are as follows:

[0146]

[0147]

[0148] Where: R' a , R' b are the initial threshold radius estimated according to the target’s speed, are the predicted speeds at time k (based on time k-1) in the xy direction, α and β are the magnification coefficients, and Thrd max The maximum value of the threshold radius. When the estimated threshold radius exceeds this value, it is set to Thrd max ,Thrd min It is the minimum value of the threshold radius. When the estimated threshold radius is 0 or less than this value, it is set to Thrd min .

[0149] Among them, the calculation and update of α and β at time k can be based on the errors between the predicted position and the actual measured position at time k-1 and time k-2, and the magnification value at the current time k can be obtained by using the chi-square test.

[0150]

[0151]

[0152]

[0153] in: They are the predicted positions in the xy direction at time k-1 (based on time k-2);

[0154] m x(k-1) 、m y(k-1) The distribution is the measured position in the xy direction at time k-1;

[0155] Std_x and Std_y are the standardized values ​​of the chi-square test in the xy direction, respectively;

[0156] are the normalized values ​​in the x and y directions respectively;

[0157] ChiSquaredCdf is the chi-square test query probability table.

[0158] The present invention designs an adaptive ellipse association gate model, combines the motion state and speed of the tracking target, designs an ellipse model whose major axis always keeps consistent with the motion direction, calculates the radii of the major and minor axes of the ellipse through the chi-square test, adaptively adjusts the direction and size of the ellipse association model, reduces unnecessary area calculations, and reduces the amount of algorithm computation.

[0159] Step S2.2, determining the measurement target falling into the associated wave gate.

[0160] For the tracking target M = {m i , m2, ..., m i ,}, the current frame measurement target N = {n1, n2, ..., n j ,}, the correlation wave gate model constructed in step S2.1 is brought into the calculation to determine whether the measurement target falls into the correlation wave gate, and a preliminary correlation candidate matrix A is obtained, where 1 in the matrix represents falling into the correlation wave gate, and 0 represents not falling into the correlation wave gate.

[0161]

[0162] Step S3, calculating the association probability between the tracking target and the measurement target within the association gate, and setting the association probability to 0 for the measurement target that does not fall into the association gate, thereby obtaining an association probability matrix.

[0163] The existing technology only considers the association probability matrix calculation algorithm of single-dimensional information, which is easy to cause target association errors, target association failures, etc., resulting in target tracking ID jumps and affecting target tracking stability.

[0164] The association probability of the present invention is an association probability weighted based on multi-dimensional information. In this embodiment, the weighting can be performed by, but is not limited to, using target detection probability score, IOU intersection-over-union probability, and Euclidean distance probability.

[0165] The target detection probability score is a known variable.

[0166] For n measurement targets Mea={m1, m2, ..., m n},m i =(x m|i ,y m|i ) and the jth tracking target T j The predicted position (x t|j ,y t|j ) Calculate the IOU and Euclidean distance to get the following set:

[0167] IOU={ikou 1,j ,iou 2,j ,...,iou n,j} Formula (21)

[0168] Eu_Distance = {eu 1,j ,eu 2,j ,...,eu n,j} Formula (22)

[0169] Among them, Figure 8 As shown, the calculation formula of IOU is:

[0170]

[0171] The calculation formula of Euclidean distance is:

[0172] eu i,j =((x m|i -x t|j ,) 2 +(y m|i -y t|j ,) 2 ) 1 / 2 Formula (24)

[0173] Because the value of the Euclidean distance is not completely between (0, 1), the Euclidean distance needs to be normalized. The normalization formula is shown in Formula 25:

[0174]

[0175] Calculate the weighted probability:

[0176] P w|i,j =α*S i,j +β*ioui,j +γ*eu n,j ', stα+β+γ=1 Formula (26)

[0177] Among them: α, β, γ are weighted indexes, which are empirical values. The size of the value represents whether a certain variable is more trusted.

[0178] For example, α, β, and γ are 0.5, 0.3, and 0.2 respectively, indicating that the weight of the confidence in the detection is 0.5 and the IOU is 0.3. w|i,j It needs to be normalized to the interval (0, 1) and needs to be normalized using formula (25) (replace Eu_Distance with P w ).

[0179] The present invention optimizes the association probability calculation of the association probability matrix, combines the target detection probability score, IOU and Euclidean distance, etc., and obtains the weighted probability by weighted average to replace the single detection probability, thereby improving the reliability of association in multiple dimensions, improving the association accuracy between targets, and thus improving the tracking stability.

[0180] Step S4, based on the association probability matrix between the tracking target and the measurement target, the Hungarian algorithm is used to calculate the distribution relationship between the tracking target and the measurement target.

[0181] Step S5: recalculate the Euclidean distances between the unassigned tracking targets and measurement targets, generate a new association probability matrix, and use the Hungarian algorithm to calculate a new assignment relationship. Unassociated targets are associated through the minimum distance.

[0182] Due to the influence of detection error, the association probability calculation may be small and fail to reach the association threshold, resulting in association failure. In order to reduce the impact of detection error and improve the association accuracy of the target, the Euclidean distance is used for the final association matching.

[0183] For the unmatched tracking target U_M={u_m1,u_m2,...,u_m i ,}, the current frame measurement target U_N = {u_n1, u_n2, ..., u_n j ,}, calculate the Euclidean distance between each pair (refer to Formula 20 for the Euclidean distance formula) and obtain a new association probability matrix A.

[0184]

[0185] For the association probability matrix A, the Hungarian matching algorithm is also used to allocate the association probability matrix to obtain the allocation relationship between the tracking target and the measurement target.

[0186] Step S6: updating the tracking target motion state and Kalman motion filter model of the tracking target according to the latest allocation relationship.

[0187] For the tracking target associated with the measurement at time t, the filter position and covariance matrix are updated through the Kalman update equation of step S2. For the measurement not associated with any tracking target, a new trajectory is created for management and tracking. For the unassociated tracking target, at time t, only its position state and covariance state matrix are predicted according to the Kalman prediction equation of step S2.

[0188] In summary, the present invention firstly derives the predicted position at time t through the tracking target motion model at time t-1, and calculates the associated wave gate at time t in combination with the tracking target motion state; then, based on the improved weighted probability calculation method, the associated probability is calculated for the detected target falling into the associated wave gate, and sent to the target filtering model to update the target motion state at time t, thereby completing the target tracking from time t-1 to time t.

[0189] In some other embodiments, the present invention discloses a target tracking device based on an adaptive correlation gate, comprising:

[0190] A tracking target determination module is used to determine the tracking target motion state and filtering model according to the tracking target type;

[0191] A correlation wave gate determination module is used to determine the correlation wave gate according to the tracking target type and the tracking target motion state;

[0192] The association probability calculation module is used to calculate the association probability between the tracking target and the measurement target within the association wave gate. For the measurement target that does not fall into the association wave gate, the association probability is set to 0 to obtain the association probability matrix;

[0193] A distribution relationship preliminary calculation module is used to calculate the distribution relationship between the tracking target and the measurement target using the Hungarian algorithm based on the association probability matrix between the tracking target and the measurement target;

[0194] The allocation relationship secondary calculation module is used to recalculate the Euclidean distance between unallocated tracking targets and measurement targets, generate a new association probability matrix, and calculate the new allocation relationship using the Hungarian algorithm;

[0195] The updating module is used to update the tracking target motion state and the filtering model of the tracking target according to the latest allocation relationship.

[0196] Further, the associated wave gate determination module includes:

[0197] A correlation wave gate model determination unit, used to determine the correlation wave gate model according to the tracking target type and the tracking target motion state;

[0198] If the tracking target type is a stationary target, or the tracking target type is a moving target and the moving state of the tracking target is stationary, a circular correlation wave gate of a fixed size is selected to obtain a correlation wave gate model;

[0199] If the tracking target type is a moving target and the tracking target motion state is moving, then the elliptical correlation wave gate that conforms to the current tracking target motion state is calculated to obtain a correlation wave gate model;

[0200] The judgment unit is used to bring the measurement target set of the current frame into the correlation wave gate model to judge whether it falls into the correlation wave gate.

[0201] Furthermore, the correlation wave gate determination module further includes: an elliptical correlation wave gate calculation unit, which is used to calculate an elliptical correlation wave gate that conforms to the current tracking target motion state, specifically including:

[0202] A rotation direction calculation unit is used to calculate the direction angle of the tracking target's motion speed based on the tracking target's motion state, and obtain an elliptical correlation wave gate with an adaptive rotation direction, wherein the major axis direction of the elliptical correlation wave gate is consistent with the speed direction of the tracking target;

[0203] The radius calculation unit is used to calculate the major and minor axis radii of the elliptical correlation wave gate according to the motion state of the tracking target.

[0204] Furthermore, in the association probability calculation module, the association probability is a weighted association probability based on multi-dimensional information.

[0205] In this embodiment, the specific content of the target tracking device based on the adaptive correlation wave gate is similar to the content of the target tracking method based on the adaptive correlation wave gate disclosed in the above embodiment, and will not be repeated here.

[0206] An embodiment of the present invention discloses a storage medium storing one or more computer-readable programs, wherein the one or more programs include instructions suitable for being loaded by a memory and executing the target tracking method based on adaptive correlation gate disclosed in the above embodiment.

[0207] The nearest neighbor algorithm, probability density association algorithm, and joint probability density association algorithm mentioned in the background technology all involve the design of association gates and the calculation of association probability matrices. However, fixed-size association gates and single-dimensional association cost calculation methods are usually used, which may cause target association failure in some cases, resulting in target tracking loss.

[0208] In view of the above situation, the present invention proposes a target tracking method, device and storage medium based on adaptive correlation gate, which optimizes the correlation gate by tracking the target motion state, screens the target and reduces the algorithm calculation amount; and optimizes the association probability calculation method, uses multiple dimensions (such as detection confidence, IOU, Euclidean distance, etc.) to calculate the weighted association probability, improves the target association accuracy, and thus improves the stability of target tracking.

[0209] In summary, the present invention discloses a target tracking method, device and storage medium based on adaptive correlation gate, which has the following beneficial effects:

[0210] First, the present invention adaptively adjusts the size of the correlation gate according to the motion state of the tracked target, reducing the number of association failures caused by data changes in the fixed-size correlation gate. Further, the direction and size of the actual elliptical correlation gate are adjusted by tracking the motion state of the target, which is used to filter the measured target, realize the association between the tracked predicted target and the measured target, improve the accuracy of target association, improve the target tracking effect, and reduce the safety risks caused by it; including but not limited to other clustering or association algorithms that can be designed as elliptical or circular correlation gate models, such as the nearest neighbor algorithm (NN), probability density estimation (PDA), joint probability density estimation (JPDA), etc., should all fall within the scope of protection.

[0211] Second, the present invention uses weighted association probability to calculate association probability and obtain the distribution relationship between the tracking target and the measurement target. By calculating the weighted association probability, the association correlation is considered in multiple dimensions, which reduces the limitations of a single dimension, improves the accuracy of target association, and thus reduces the probability of target loss and improves the stability of target tracking. Including but not limited to algorithms that use probability to calculate target association, such as probability density estimation (PDA), joint probability density estimation (JPDA), multi-target hypothesis (MHT), etc., should all fall within the scope of protection.

[0212] The present invention can be applied in the field of laser radar to achieve target association tracking optimization based on laser radar. Other sensor target association schemes involving IOU and distance association, such as millimeter wave, radar, camera, etc., all fall within the scope of protection.

[0213] Of course, the present invention can also be applied in the fields of autonomous driving, traffic radar, security monitoring, etc. The present invention can improve the accuracy of target association and thus improve the stability of target tracking.

[0214] It should be understood that the various techniques described herein can be implemented in combination with hardware or software, or a combination thereof. Thus, the method and apparatus of the present invention, or some aspects or portions of the method and apparatus of the present invention, can take the form of program codes (i.e., instructions) embedded in a tangible medium, such as a floppy disk, a CD-ROM, a hard disk drive, or any other machine-readable storage medium, wherein when the program is loaded into a machine such as a computer and executed by the machine, the machine becomes a device for practicing the present invention.

[0215] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for illustrating the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which shall fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. Target tracking method based on adaptive correlation gate, It is characterized in that include: Step S1: Determine the motion state and filtering model of the tracking target according to the tracking target type; Step S2: determining the associated wave gate according to the tracking target type and the tracking target motion state; Step S3: Calculate the association probability between the tracking target and the measurement target within the association wave gate. For the measurement target that does not fall into the association wave gate, the association probability is set to 0, and the association probability matrix is ​​obtained; Step S4: calculating the distribution relationship between the tracking target and the measurement target based on the association probability matrix between the tracking target and the measurement target; Step S5: for the unassigned tracking targets and measurement targets, recalculate the Euclidean distances between them, generate a new association probability matrix, and calculate a new assignment relationship; Step S6: According to the latest allocation relationship, the tracking target motion state and the filter model of the tracking target are updated.

2. The target tracking method according to claim 1, It is characterized in that The step S2 comprises: Step S2.1: determining the correlation gate model according to the tracking target type and the tracking target motion state; If the tracking target type is a stationary target, or the tracking target type is a moving target and the moving state of the tracking target is stationary, a circular correlation wave gate of a fixed size is selected to obtain a correlation wave gate model; If the tracking target type is a moving target and the tracking target motion state is moving, then the elliptical correlation wave gate that conforms to the current tracking target motion state is calculated to obtain a correlation wave gate model; Step S2.2: Bring the measurement target set of the current frame into the correlation gate model to determine whether it falls into the correlation gate.

3. The target tracking method according to claim 2, It is characterized in that The following steps are used to calculate the elliptical correlation gate that matches the current tracking target motion state: Step A: Based on the motion state of the tracking target, the orientation angle of the motion speed of the tracking target is calculated to obtain an elliptical correlation wave gate with an adaptive rotation direction, and the major axis direction of the elliptical correlation wave gate is consistent with the velocity direction of the tracking target; Step B: Calculate the major and minor axis radii of the elliptical correlation wave gate according to the motion state of the tracking target.

4. The target tracking method according to claim 1, It is characterized in that In step S3, the association probability is a weighted association probability based on multi-dimensional information.

5. Target tracking device based on adaptive correlation gate, It is characterized in that include: A tracking target determination module is used to determine the tracking target motion state and filtering model according to the tracking target type; A correlation wave gate determination module is used to determine the correlation wave gate according to the tracking target type and the tracking target motion state; The association probability calculation module is used to calculate the association probability between the tracking target and the measurement target within the association wave gate. For the measurement target that does not fall into the association wave gate, the association probability is set to 0 to obtain the association probability matrix; A distribution relationship preliminary calculation module, used for calculating the distribution relationship between the tracking target and the measurement target based on the association probability matrix between the tracking target and the measurement target; The allocation relationship secondary calculation module is used to recalculate the Euclidean distance between unallocated tracking targets and measurement targets, generate a new association probability matrix, and calculate a new allocation relationship; The updating module is used to update the tracking target motion state and the filtering model of the tracking target according to the latest allocation relationship.

6. The target tracking method according to claim 5, It is characterized in that The associated wave gate determination module comprises: A correlation wave gate model determination unit, used to determine the correlation wave gate model according to the tracking target type and the tracking target motion state; If the tracking target type is a stationary target, or the tracking target type is a moving target and the moving state of the tracking target is stationary, a circular correlation wave gate of a fixed size is selected to obtain a correlation wave gate model; If the tracking target type is a moving target and the tracking target motion state is moving, then the elliptical correlation wave gate that conforms to the current tracking target motion state is calculated to obtain a correlation wave gate model; The judgment unit is used to bring the measurement target set of the current frame into the correlation wave gate model to judge whether it falls into the correlation wave gate.

7. The target tracking device according to claim 6, It is characterized in that The correlation wave gate determination module further includes: an elliptical correlation wave gate calculation unit, which is used to calculate an elliptical correlation wave gate that conforms to the current tracking target motion state, specifically including: A rotation direction calculation unit is used to calculate the direction angle of the tracking target's motion speed based on the tracking target's motion state, and obtain an elliptical correlation wave gate with an adaptive rotation direction, wherein the major axis direction of the elliptical correlation wave gate is consistent with the speed direction of the tracking target; The radius calculation unit is used to calculate the major and minor axis radii of the elliptical correlation wave gate according to the motion state of the tracking target.

8. The target tracking device according to claim 5, It is characterized in that In the association probability calculation module, the association probability is a weighted association probability based on multi-dimensional information.

9. Storage media, It is characterized in that The storage medium stores one or more computer-readable programs, and the one or more programs include instructions, and the instructions are suitable for being loaded by the memory and executing the target tracking method based on adaptive correlation gate as described in any one of claims 1-4.

Citation Information

Cited By

  • Improved adaptive tracking gate target tracking method

    CN119959927A

  • An improved adaptive tracking door target tracking method

    CN119959927B

  • Target tracking method of low-altitude strong-maneuvering unmanned aerial vehicle

    CN121385867A