A trajectory tracking method with adaptive unknown detection probability for multi-type target scenes
Patent Information
- Application Number
- CN202610788838.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-03
- Publication Date
- 2026-09-25
AI Technical Summary
[0004]本发明的目的在于提供一种多类型目标场景自适应未知检测概率的轨迹跟踪方法，旨在解决点扩展目标共存场景下跟踪精度低的问题
[0004]本发明的目的在于提供一种多类型目标场景自适应未知检测概率的轨迹跟踪方法，旨在解决点扩展目标共存场景下跟踪精度低的问题。
Smart Images
Figure CN122815409A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar multi-target tracking technology, specifically to a trajectory tracking method with adaptive unknown detection probability for multi-type target scenes. Background Technology
[0002] While random finite set (RFS) theory avoids explicit data association in multi-target tracking by describing the state of multiple targets through sets, most assumptions rely on fixed and known detection probabilities. Furthermore, it cannot simultaneously output target trajectories and distinguish between point targets and extended targets, leading to insufficient tracking accuracy in mixed scenes. Therefore, a trajectory tracking method with adaptive unknown detection probabilities for multi-type target scenes is proposed based on trajectory Poisson-Dobernouri mixture (TPMBM) of random finite sets.
[0003] For multi-type targets, research on the conjugate prior of TMBM under unknown detection probabilities has not yet been achieved. Summary of the Invention
[0004] The purpose of this invention is to provide an adaptive trajectory tracking method with unknown detection probabilities for multi-type target scenes, aiming to solve the problem of low tracking accuracy in scenarios where point-extended targets coexist.
[0005] To achieve the above objectives, this invention provides a trajectory tracking method with adaptive unknown detection probability for multi-type target scenes, comprising the following steps:
[0006] Step 1: Define the detection probability as a random variable and construct a generalized augmented trajectory space to accommodate point targets and extended target trajectories;
[0007] Step 2: Model the unknown detection probability using the beta distribution, and represent the target trajectory state as a weighted mixture of the beta-gamma-gaussian inverse Westerly distribution and the beta-gaussian distribution;
[0008] Step 3: Predict the Poisson-to-Bernoulli mixture of the survival trajectories to obtain the predicted density of the Poisson-to-Bernoulli mixture of the survival trajectories;
[0009] Step 4: Update the Poisson-Bernoulli mixture of the predicted survival trajectories to obtain the updated density of the Poisson-Bernoulli mixture of survival trajectories;
[0010] Step 5: Filter to obtain the trajectory of surviving targets, and introduce a detection probability merging strategy to improve the accuracy of detection probability estimation.
[0011] Optionally, in step 1, the generalized augmented trajectory space is defined as follows: The trajectory space of the point target and expanding the target trajectory space All include birth time, survival time, state sequence, and detection probability sequence; the generalized augmented trajectory is represented as ,in For the time of birth, For survival time, It is a state sequence. For detecting probability sequences; the type of the generalized augmented trajectory is determined by the state sequence. The type is determined.
[0012] Optionally, the detection probability in step 2 Using beta distribution Modeling, in which and The distribution parameters are: the point target trajectory state is modeled using a Gaussian distribution, the extended target trajectory state is modeled using a gamma-Gaussian inverse Westerly distribution, and the weighted mixture form is represented as B-GGIWG.
[0013] Optionally, the prediction step in step 3 includes calculating the prediction intensity of the Poisson process for the survival trajectory. The predicted probability density is obtained by mixing with the Bernoulli model, where the prediction of the survival trajectory is based on the state transition model and the detection probability transition model; the intensity of the newborn trajectory is given by the prior birth distribution.
[0014] Optionally, in step 4, the update step includes: for a given measurement set, processing undetected updates and detected updates separately; undetected updates update the trajectory probability density using only an empty measurement set, while detected updates calculate the likelihood function using a non-empty measurement set and adjust the target type probability. To distinguish between point targets and extended targets.
[0015] Optionally, in step 5, the detection probability merging strategy is expressed as follows: ,in For the target total number, For the first The estimated detection probability of each target is calculated; during state extraction, a probability threshold is set, and only the trajectory of targets with a probability higher than the threshold is output.
[0016] This invention provides a trajectory tracking method with adaptive unknown detection probabilities for multi-type target scenarios. First, the detection probability is defined as a random variable and augmented to the target trajectory state space, constructing a generalized augmented trajectory space to accommodate point targets and extended targets. Second, a beta distribution is used to model the unknown detection probability, and the target trajectory state is represented as a weighted mixture of a beta-gamma-Gaussian inverse Westerly distribution and a beta-Gaussian distribution, achieving target type discrimination, dynamic adjustment of detection probabilities, and synchronous updating of trajectory states. Furthermore, a detection probability merging strategy is introduced, improving estimation accuracy through averaging. Simulation experiments show that, regardless of the target's form in a monitoring scenario, the method of this invention can accurately estimate the target's position. Extensive experiments have verified the effectiveness and robustness of this method in several challenging tracking scenarios. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart illustrating an adaptive trajectory tracking method for multi-type target scenes with unknown detection probabilities according to the present invention.
[0019] Figure 2 This is a true trajectory map of a small sample of point targets and an extended target in scenario 1 of a specific embodiment of the present invention.
[0020] Figure 3 This is a trajectory error diagram in scenario 1 of a specific embodiment of the present invention.
[0021] Figure 4 This is a detection probability estimation diagram in scenario 1 of a specific embodiment of the present invention.
[0022] Figure 5 This is a real trajectory map of five randomly generated extended targets in scenario 2 of a specific embodiment of the present invention.
[0023] Figure 6 This is the potential estimation diagram in scenario 2 of a specific embodiment of the present invention.
[0024] Figure 7 This is a detection probability estimation diagram in scenario 2 of a specific embodiment of the present invention. Detailed Implementation
[0025] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.
[0026] Please see Figure 1 This invention provides a trajectory tracking method with adaptive unknown detection probability for multi-type target scenes, comprising the following steps:
[0027] Step 1: Define the detection probability as a random variable and construct a generalized augmented trajectory space to accommodate point targets and extended target trajectories;
[0028] Step 2: Model the unknown detection probability using the beta distribution, and represent the target trajectory state as a weighted mixture of the beta-gamma-gaussian inverse Westerly distribution and the beta-gaussian distribution;
[0029] Step 3: Predict the Poisson-to-Bernoulli mixture of the survival trajectories to obtain the predicted density of the Poisson-to-Bernoulli mixture of the survival trajectories;
[0030] Step 4: Update the Poisson-Bernoulli mixture of the predicted survival trajectories to obtain the updated density of the Poisson-Bernoulli mixture of survival trajectories;
[0031] Step 5: Filter to obtain the trajectory of surviving targets, and introduce a detection probability merging strategy to improve the accuracy of detection probability estimation.
[0032] The following provides further explanation with reference to the specific steps:
[0033] Step 1: Define the detection probability as a random variable and construct a generalized augmented trajectory space to accommodate point targets and extended target trajectories.
[0034] During execution, the process also includes merging the unknown detection probability space with the point target trajectory state space and the extended target trajectory state space to form an augmented trajectory state space.
[0035] Specifically, the trajectory state of a single target is represented as ,in The moment when the trajectory begins. For the trajectory length, Indicates the trajectory length is The target state sequence; based on the characteristics of the detection probability, it can be defined as a real-valued interval. Real-valued variables within Detection probability sequence Belonging to space Furthermore, the detection probability is extended to the trajectory space of both point targets and extended targets. A generalized augmented trajectory space capable of accommodating both point targets and extended targets is constructed through direct summation operations. Based on this, the generalized augmented trajectory space model is expressed as: ;
[0036]
[0037] Augmented trajectory ,like ,but The augmented trajectory of the representative point target; if ,but This is represented as the augmented trajectory of the extended target. At any given time, assuming there is a reason A set of augmented trajectories ,here It is an augmented trajectory space A random finite subset.
[0038] Step 2: Model the unknown detection probability using the beta distribution, and represent the target trajectory state as a weighted mixture of the beta-gamma-gaussian inverse Westerly distribution and the beta-gaussian distribution.
[0039] Specifically, the beta distribution is used to model the probability of unknown detection, and the gamma-gaussian inverse Westerly (GGIW) distribution and Gaussian distribution are used to describe the trajectory states of extended targets and point targets.
[0040] During execution, the unknown detection probability at the current moment is also combined with the trajectory states of the extended target and the point target, and expressed as a beta-gamma Gaussian inverse Westerly distribution (BGGIW) and a beta-gaussian distribution (BG), respectively.
[0041] Specifically regarding augmented trajectories In this invention, only the detection probability at the current moment is considered. By modeling with Bethesda distribution, the following conditions are met: , The details are as follows:
[0042]
[0043] The mean and covariance of its distribution are respectively , ,in This represents the beta function.
[0044] Assuming in Augmented trajectory of the point of birth at any given moment Its unknown detection probability The trajectory state is determined by the mean sequence. With covariance sequence The Gaussian distribution representation is as follows:
[0045]
[0046] in .
[0047] Assuming in Extended target augmentation trajectory of birth moment Its unknown detection probability The trajectory state is modeled using a gamma-Gaussian inverse Westerly distribution (GGIW). Its target measurement generation rate is modeled by a gamma distribution derived from the parameter sequence. , Control, the target state is modeled using a Gaussian distribution, derived from the parameter sequence. This indicates that the target shape is determined by a sequence of parameters. The decision is as follows:
[0048]
[0049] in , .
[0050] Step 3: Predict the Poisson-Bernoulli mixture of the survival trajectories to obtain the predicted density of the survival trajectory Poisson-Bernoulli mixture.
[0051] The execution process includes predicting the newborn trajectory, potential trajectory, and BGGIW and BG mixture assumptions of the survival trajectory in the Poisson-Do-Bernoulli mixture of the survival trajectory.
[0052] Specifically, assuming the birth intensity of the target augmenting trajectory is a mixture of BGGIW and BG, denoted as:
[0053]
[0054] in , These represent the number of birth intensity components (BG and BGGIW), respectively. , They represent the first The weights of each component and the time of birth.
[0055] Assuming in At that moment, for the first Bernoulli's first The augmented trajectory density of each local hypothesis is:
[0056]
[0057] Here and These represent the probabilities that the augmented trajectory is a point target trajectory and the extended target trajectory, respectively.
[0058] Assume that at time k, the trajectory parameters of the extended target and the point target are given by... , , , Confirmed. To estimate the trajectory state, the parameters are predicted by augmenting and expanding the GGIW prediction of the target trajectory and using Kalman filtering, as shown below:
[0059]
[0060]
[0061] The prediction parameters for the extended target and the point target in the augmented trajectory are: , At the same time .
[0062] The prediction of potential trajectory targets and emerging trajectories follows the formula below:
[0063]
[0064] Where parameters , , , and , where is the prediction parameter.
[0065] For MBM density prediction of augmented trajectories, the overall assumptions remain unchanged. In the MBM's... Bernoulli has , , When predicting the trajectory state, the trajectory is based on... The probability of Bernoulli surviving. Therefore, the predicted probability of Bernoulli's existence. Bernoulli density is given by the following formula:
[0066]
[0067] This formula, through different weights and The corresponding BG components and BGGIW components are merged. , .
[0068] Step 4: Update the Poisson-Bernoulli mixture of the predicted survival trajectories to obtain the updated density of the survival trajectory Poisson-Bernoulli mixture.
[0069] The execution process includes updating the predicted new trajectories, potential trajectories, the multi-Bernoulli mixture components of the trajectories, and the probability of their existence. Specifically, the predicted trajectory density is the same as in step S3. The probability density updates of the Poisson-Bernoulli mixture at time step are as follows:
[0070] 1) Undetected updates to the PPP strength of potential augmented trajectories:
[0071]
[0072] Among them, the time parameter The updated augmented trajectory extension target parameters can be divided into two forms: one is the case where no detection is caused by the detection probability, and the other is the case where no measurement is generated due to the measurement rate, as follows:
[0073]
[0074] The updated augmented trajectory point target parameters are as follows:
[0075]
[0076] in, and The difference represents the detection probability of augmented trajectory extended targets and point targets.
[0077] 2) No. A trajectory Undetected updates of the augmented trajectory Bernoulli components of the hypothetical cases:
[0078]
[0079] in , , ,
[0080]
[0081] The updated augmented trajectory expansion target parameters are as follows:
[0082]
[0083] The updated augmented trajectory point target trajectory parameters are as follows:
[0084]
[0085] 3) Given a non-empty measurement subset at time k+1 When, for the i-th Bernoulli component, the first... The assumptions are updated. We can assume that this Bernoulli exists, therefore its probability of existence is... At this point, the relevant parameter update formula is as follows:
[0086]
[0087] like Point target trajectory type probability Then the remaining parameters are updated as follows:
[0088]
[0089] in, and These are the augmented trajectory expansion target parameters and the likelihood function, respectively.
[0090] like Then we have:
[0091]
[0092] in, and These are the target parameters and likelihood function of the augmented trajectory points, respectively.
[0093] 4) Given a non-empty subset of measurements at time k+1 When updating the PPP strength, a new Bernoulli component will be generated. hour , Therefore, we only consider this. Updates to the first detected potential trajectory:
[0094]
[0095] Here , , ,in , as well as , These are the updated parameters.
[0096] if Then there must exist an extended target trajectory, such that , At this point, we have:
[0097]
[0098] if ,at this time Includes only one measurement value Then we have:
[0099]
[0100] Step 5: Filter to obtain the trajectory of surviving targets, and introduce a detection probability merging strategy to improve the accuracy of detection probability estimation.
[0101] The execution process includes extracting the BGGIWG components of the trajectory Bernoulli components with a probability of 1 in the trajectory Bernoulli mixture after the filter at time k, obtaining the BGGIWG components of the trajectories from time 1 to time k, and then applying the type hypothesis function. The target type is identified and the target trajectory is extracted, yielding the target trajectory detection probability. Subsequently, to obtain a more stable detection probability, this paper merges the estimated detection probabilities for each target, letting... This background detection probability is used in subsequent tracking experiments. This indicates the total number of targets in the scene. Indicates the first The detection probability of each target.
[0102] Further, please refer to Figures 2 to 7 The present invention also proposes a specific embodiment, which is further illustrated through simulation experiments:
[0103] 1. Simulation conditions: This invention was tested on a device with a 12th Gen Intel(R) Core(TM) i7-12700 (2.10 GHz) and 16.0 GB of baseband RAM, and the simulation was completed using MATLAB R2021b software.
[0104] 2. Simulation scene setup: Considering in In a two-dimensional scene, the clutter Poisson ratio Set different detection probabilities and with The threshold for distinguishing point targets from extended targets, when When the target is at a certain point, the trajectory is taken as the point target; otherwise, the extended target is taken. Transition matrix. and process noise They are represented as follows:
[0105]
[0106] in It is the sampling interval. It is a second-order identity matrix. It is the standard deviation in process noise.
[0107] The intensity of the newborn distribution is modeled as a mixture of BGGIW and BG, where the birth intensity of the target trajectory is represented. , , Target trajectory parameters , The state parameters of the point target and the extended target are described as follows:
[0108]
[0109] In scenario 1, four randomly generated targets move through the monitored area for 100 time steps using a uniform motion model, such as... Figure 2 As shown. In scenario 1, the trajectory error diagram is as follows. Figure 3 As shown, initially, there is an initial cost due to the failure to detect the target in time; as the target is successfully captured, the error decreases. At 20s and 35s, the introduction of new targets causes a phased increase in the error of this invention. Overall, this invention consistently maintains the lowest error level, indicating that it possesses optimal performance. Different detection probability estimation graphs are shown below. Figure 4 As shown, the estimated detection probability exhibits a temporary decrease at 20s and 35s. This is because the initial detection probability of newly formed targets is low, causing the estimated value to be lower than the true mean. Similarly, at 80s, target disappearance is misjudged as a missed detection, causing a brief drop in the estimated value before it stabilizes. Experiments demonstrate that this invention can still achieve accurate identification of detection probabilities even under conditions of unknown detection probabilities.
[0110] In scenario 2, there are 5 randomly generated targets that move through the monitored area for 100 time steps using a uniform motion model, such as... Figure 5 As shown. The clutter rate and detection probability in scenario 2 are the same as in scenario 1.
[0111] In scenario 2, the potential estimation graph is as follows: Figure 6 As shown, the solution provided by this invention can accurately estimate the number of targets even when the detection probability of a single type of target is unknown. However, it still has an insurmountable problem: a certain lag still occurs when the target trajectory disappears. The sensor detection probability estimation diagram is shown below. Figure 7 As shown, Figure 7 As shown, the red curve represented by the present invention is closer to the true value, indicating that the present invention's estimation of the unknown detection probability is more accurate and stable. However, at time step 78, the estimated detection probability drops sharply. This drop is attributed to the significant impact of target disappearance. Furthermore, different detection probabilities have different effects on the identification of target disappearance. Higher detection probabilities enhance the ability to identify target disappearance and can confirm target death more quickly.
[0112] In summary, simulation results demonstrate that the present invention can accurately estimate the target location regardless of the type or location of the target in the monitoring scenario. Extensive experiments have verified the effectiveness and robustness of the present invention in multiple challenging tracking scenarios.
[0113] The above description discloses only one preferred embodiment of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art will understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.
Claims
1. A trajectory tracking method with adaptive unknown detection probability for multi-type target scenes, characterized in that, Includes the following steps: Step 1: Define the detection probability as a random variable and construct a generalized augmented trajectory space to accommodate point targets and extended target trajectories; Step 2: Model the unknown detection probability using the beta distribution, and represent the target trajectory state as a weighted mixture of the beta-gamma-gaussian inverse Westerly distribution and the beta-gaussian distribution; Step 3: Predict the Poisson-to-Bernoulli mixture of the survival trajectories to obtain the predicted density of the Poisson-to-Bernoulli mixture of the survival trajectories; Step 4: Update the Poisson-Bernoulli mixture of the predicted survival trajectories to obtain the updated density of the Poisson-Bernoulli mixture of survival trajectories; Step 5: Filter to obtain the trajectory of surviving targets, and introduce a detection probability merging strategy to improve the accuracy of detection probability estimation.
2. The trajectory tracking method for adaptive unknown detection probability in multi-type target scenes as described in claim 1, characterized in that, In step 1, the generalized augmented trajectory space is defined as follows: The trajectory space of the point target and expanding the target trajectory space All include birth time, survival time, state sequence, and detection probability sequence; the generalized augmented trajectory is represented as ,in For the time of birth, For survival time, It is a state sequence. For detecting probability sequences; the type of the generalized augmented trajectory is determined by the state sequence. The type is determined.
3. The trajectory tracking method for adaptive unknown detection probability in multi-type target scenes as described in claim 2, characterized in that, Detection probability in step 2 Using beta distribution Modeling, in which and The distribution parameters are: the point target trajectory state is modeled using a Gaussian distribution, the extended target trajectory state is modeled using a gamma-Gaussian inverse Westerly distribution, and the weighted mixture form is represented as B-GGIWG.
4. The trajectory tracking method for adaptive unknown detection probability in multi-type target scenes as described in claim 3, characterized in that, The prediction step in step 3 includes calculating the prediction strength of the Poisson process for the survival trajectory. The predicted probability density is obtained by mixing with the Bernoulli model, where the prediction of the survival trajectory is based on the state transition model and the detection probability transition model; the intensity of the newborn trajectory is given by the prior birth distribution.
5. The trajectory tracking method for adaptive unknown detection probability in multi-type target scenes as described in claim 4, characterized in that, In step 4, the update steps include: for a given measurement set, processing undetected updates and detected updates separately; undetected updates update the trajectory probability density using only an empty measurement set, while detected updates calculate the likelihood function using a non-empty measurement set and adjust the target type probability. To distinguish between point targets and extended targets.
6. The trajectory tracking method for adaptive unknown detection probability in multi-type target scenes as described in claim 5, characterized in that, In step 5, the detection probability merging strategy is expressed as follows: ,in For the target total number, For the first The estimated detection probability of each target is calculated; during state extraction, a probability threshold is set, and only the trajectory of targets with a probability higher than the threshold is output.