Multi-target tracking method, system and device based on compressed sensing and particle filtering and medium

Abstract

The invention discloses a multi-target tracking method and system based on compressed sensing and particle filtering, and a medium, and belongs to the field of signal monitoring and tracking, and the method comprises the steps: carrying out the compression observation of a received original array signal based on the multi-target tracking of compressed sensing and particle filtering, and obtaining a low-dimensional compression observation signal; respectively initializing a group of independent particle swarms for each target to be tracked; according to the low-dimensional compressed observation signal, respectively executing time recursion tracking on the particle swarm of each to-be-tracked target, and jointly calculating the weight of each particle in each particle swarm to obtain a weighted particle set of each to-be-tracked target; state estimation is carried out according to the weighted particle set of each to-be-tracked target, an angle estimation value and an angular velocity estimation value of each to-be-tracked target are obtained, and a tracking result of each to-be-tracked target is output. And real-time combined tracking of angles and speeds of a plurality of to-be-tracked targets is realized.

Description

Technical Field

[0001] This invention relates to the field of signal monitoring and tracking, and in particular to a multi-target tracking method, system, device and medium based on compressed sensing and particle filtering. Background Technology

[0002] Currently, multi-target angle tracking technology mainly relies on array signal processing-based methods such as traditional beamforming, the MUSIC algorithm, and the ESPRIT algorithm. These methods typically require complete array reception data, have high computational complexity, and demand high signal-to-noise ratio and target separation angle. In recent years, compressed sensing technology has played an important role in angle estimation. Compressed sensing can alleviate the requirements for high signal-to-noise ratio and multiple snapshots by reducing the sampling dimension, thereby reducing hardware costs and computational burden. However, it still suffers from problems such as insufficient tracking capability for dynamic targets, inaccurate velocity estimation, and non-robust particle filter initialization.

[0003] Existing multi-target tracking methods suffer from the following drawbacks: 1. High-dimensional data dependence: Traditional methods require the construction of a signal covariance matrix based on data received from a complete array, resulting in high hardware costs and computational complexity. 2. Weak dynamic tracking capability: Existing compressed sensing methods often design observation matrices based on static sparse signal models, failing to incorporate dynamic constraints of the target's motion state, leading to inaccurate estimation of angle change rate (velocity). 3. Inaccurate particle filter initialization: The generation of the initial particle set often relies on empirical prior distributions such as uniform or Gaussian distributions, without incorporating prior information about the target's motion, resulting in slow convergence and large tracking deviations. 4. Multi-target interference: Targets influence each other, easily leading to angle confusion or tracking loss. 5. Lack of joint estimation mechanism: Existing technologies generally separate angle estimation and velocity estimation into independent processing modules, lacking collaborative optimization. Summary of the Invention

[0004] This invention provides a multi-target tracking method, system, device, and medium based on compressed sensing and particle filtering. By compressing observations to reduce data dimensionality, it enables real-time joint tracking of the angles and velocities of multiple targets to be tracked.

[0005] This invention provides a multi-target tracking method based on compressed sensing and particle filtering, including: Compressed observations are performed on the received raw array signals to obtain low-dimensional compressed observation signals; A separate set of particle swarms is initialized for each target to be tracked; each particle in the particle swarm contains the angle state variable and angular velocity state variable of the target to be tracked; Based on the low-dimensional compressed observation signal, time recursive tracking is performed on the particle swarm of each target to be tracked, and the weight of each particle in each particle swarm is jointly calculated to obtain the weighted particle set of each target to be tracked. State estimation is performed based on the weighted particle set of each target to be tracked, and the angle estimate and angular velocity estimate of each target to be tracked are obtained. The tracking result of each target to be tracked is then output.

[0006] This invention provides a low-dimensional compressed observation signal by compressing the received raw array signal. This low-dimensional compression of data during the signal acquisition phase provides a low-dimensional data foundation for subsequent real-time data processing. A distributed state representation framework for multiple targets is established by initializing an independent particle swarm for each target, with each particle containing the target's angle and angular velocity state variables. This reduces computational complexity from exponential to linear and provides a state carrier for subsequent joint estimation of angle and velocity. Based on the low-dimensional compressed observation signal, time-recursive tracking is performed on the particle swarm, and a joint processing mechanism is used in the weight calculation stage, significantly improving the accuracy of state estimation. The angle and angular velocity estimates for each target are obtained through state estimation, and the tracking results for each target are output. This invention reduces the dimensionality burden of data processing through compressed observation and reduces the computational burden through a distributed particle swarm architecture, enabling joint tracking of the angles and velocities of multiple moving targets.

[0007] Furthermore, based on the low-dimensional compressed observation signal, time-recursive tracking is performed on the particle swarm for each target to be tracked, including: The initial process noise variance in the state prediction process is dynamically adjusted based on the error between the historical state estimate and the prior value of each target to be tracked, to obtain the first process noise variance; Based on the first process noise variance, state prediction is performed for each particle in the particle swarm of each target to be tracked, and physical range constraints are applied to obtain the constrained predicted particle state.

[0008] By introducing a closed-loop adaptive process noise adjustment mechanism into the time-recursive tracking, the process noise variance in the state prediction process is dynamically adjusted based on the error between the historical state estimate and the prior value, obtaining a first process noise variance that matches the current tracking state. The particle diffusion range is adaptively adjusted according to the actual motion error of the target, significantly enhancing the responsiveness to changes in the target's motion state and the robustness of tracking. Furthermore, based on the first process noise variance, state prediction is performed on each particle in the particle swarm, and a physically feasible range constraint is applied to the particle angular velocity after prediction, effectively preventing the particle state from spreading to an unreasonable motion range, accelerating the convergence of particles to the true state, and significantly improving the rationality of state prediction and the stability of the tracking process.

[0009] Furthermore, the weight of each particle in each particle swarm is jointly calculated to obtain a weighted particle set for each target to be tracked, specifically: Based on the predicted particle state, particles with the same index are extracted from the particle swarm of each target to be tracked to obtain an extracted particle group. A joint guidance matrix is ​​constructed based on the angular velocity state variables of each particle in the extracted particle group. The signal amplitude is estimated based on the low-dimensional compressed observation signal, the joint steering matrix, and the preset compressed observation matrix; The observation residual is calculated based on the signal amplitude. The joint basic weight of the extracted particle group is calculated by combining the predicted particle state with the signal energy penalty term. Based on the joint basic weight, the corresponding particle weight of each target to be tracked is determined, and the weighted particle set of each target to be tracked is obtained.

[0010] By extracting particles with corresponding indices from the independent particle swarms of each target based on the low-dimensional compressed observation signal, and constructing a joint steering matrix based on their angular states, accurate characterization of the observation signal is achieved under multi-target, low-dimensional observation conditions. Furthermore, based on the low-dimensional compressed observation signal, the joint steering matrix, and a preset compressed observation matrix, the signal amplitude at the prediction time is estimated. The compressed sensing reconstruction process is embedded in the particle weight evaluation, significantly improving the estimation accuracy of the signal amplitude under dimensionality reduction observation. The observation residual is calculated based on the signal amplitude, and the basic weight of the particles is calculated in conjunction with the signal energy penalty term to obtain weighted particles. This allows the residual to reflect the matching degree between the particle state and the actual observation, and the energy penalty term to suppress unreasonable estimations, thereby achieving more accurate likelihood state estimation under low-dimensional observation conditions.

[0011] Furthermore, determining the weight of each particle in the target particle swarm to be tracked, based on the joint basic weights, also includes: Based on the predicted particle state, the velocity penalty factor is calculated using the velocity penalty term formula; The joint base weights are corrected by the velocity penalty factor to determine the corresponding particle weights for each target to be tracked. The formula for the speed penalty term is as follows: in, This is the speed penalty coefficient. The range of prior angular velocities; Let be the angular velocity state variable of the i-th particle of the k-th target to be tracked in the predicted particle state at time t; K is the number of targets to be tracked; This is a speed penalty factor.

[0012] Furthermore, state estimation is performed based on the weighted particle set of each target to be tracked, obtaining the angle estimate and angular velocity estimate of each target, and the tracking result of each target is output, specifically: Based on the weighted particles, the complex average method is used to map the values ​​of the angle state variables in the weighted particles onto a complex unit circle for weighted averaging to obtain a complex result. The complex result is then converted back into an angle value to obtain an estimated angle value for the target to be tracked. Based on the weighted particles, the values ​​of the angular velocity state variables in the weighted particles are weighted and averaged to obtain a weighted average result. The weighted average result is then fine-tuned through a compensation mechanism to obtain an estimated value of the angular velocity of the target to be tracked. Based on the estimated angle and the estimated angular velocity, the tracking result for each target to be tracked is output.

[0013] By employing the complex averaging method for angle state estimation, the values ​​of the angle state variables of each particle are mapped onto a complex unit circle and weighted averaged according to the weighted particles. After obtaining the complex result, the angle value is calculated back to obtain the estimated angle value of the target to be tracked. This solves the problem of periodicity of angle variables and effectively eliminates the estimation ambiguity caused by direct linear averaging at the 360° boundary, ensuring the accuracy of angle estimation. According to the weighted particles, the values ​​of the angular velocity state variables of each particle are weighted averaged to obtain a preliminary result. Based on the historical deviation compensation mechanism, the weighted average result is fine-tuned to obtain the estimated angular velocity value of the target to be tracked. This effectively suppresses the accumulation of errors and deviations caused by model approximation or noise during long-term tracking, significantly improving the accuracy of angular velocity estimation and the overall robustness of the tracking process.

[0014] Furthermore, after obtaining the weighted particle set for each target to be tracked, intelligent resampling is also included: Based on the weighted particle set of each target to be tracked, a new particle set for each target to be tracked is generated using a system resampling method; Randomly select a portion of new particles from the new particle set, and apply differential perturbations to the state variables of the new particles to maintain particle diversity.

[0015] By performing intelligent resampling and differential perturbation before outputting the tracking results, the uniformity and representativeness of the particle set can be maintained more effectively, thereby alleviating the particle degradation problem and providing a more reliable state sample basis for subsequent tracking steps.

[0016] Furthermore, the received raw array signal is compressed and observed to obtain a low-dimensional compressed observation signal, specifically as follows: Based on the received original array signals, an original received signal vector is constructed; the original received signal vector represents the superposition of signals and noise from multiple targets to be tracked. By using a preset compressed observation matrix, the original received signal vector is linearly reduced in dimension and projected to obtain a low-dimensional compressed observation signal; wherein, the low-dimensional compressed observation signal includes a compressed observation signal vector; the number of rows of the compressed observation signal vector is much smaller than the number of array elements.

[0017] By constructing an original received signal vector containing multiple target signals and noise superposition, and then using a preset compressed observation matrix to perform linear dimensionality reduction projection on this vector, a low-dimensional compressed observation signal with a vector row count much smaller than the number of array elements is obtained. This achieves effective compression of the original array signal and significantly reduces the burden of data acquisition and transmission.

[0018] Another embodiment of the present invention provides a multi-target tracking system based on compressed sensing and particle filtering, including: a compressed observation module, an initialization particle module, a weight calculation module, and a state estimation module; The compressed observation module is used to perform compressed observation on the received original array signal to obtain a low-dimensional compressed observation signal. The initialization particle module is used to initialize an independent particle swarm for each target to be tracked; each particle in the particle swarm contains the angle state variable and angular velocity state variable of the target to be tracked. The weight calculation module is used to perform time recursive tracking of the particle swarm of each target to be tracked based on the low-dimensional compressed observation signal, and jointly calculate the weight of each particle in each particle swarm to obtain a weighted particle set of each target to be tracked. The state estimation module is used to perform state estimation based on the weighted particle set of each target to be tracked, obtain the angle estimate and angular velocity estimate of each target to be tracked, and output the tracking result of each target to be tracked.

[0019] This invention provides a low-dimensional compressed observation signal by compressing the received raw array signal. This significantly reduces the data dimensionality and transmission burden by drastically compressing the sensor data volume. A distributed state representation framework for multiple targets is established by initializing an independent particle swarm for each target, with each particle containing the target's angle and angular velocity state variables. This reduces computational complexity from exponential to linear and provides a state carrier for subsequent joint angle and velocity estimation. Based on the low-dimensional compressed observation signal, time-recursive tracking is performed on the particle swarm, and a joint processing mechanism is used in the weight calculation stage to significantly improve the accuracy of likelihood state estimation. The state estimation yields the angle and angular velocity estimates for each target, and the tracking results are output, achieving high-precision joint tracking of the angles and velocities of multiple moving targets with lower data dimensionality and computational resources.

[0020] Another embodiment of the present invention provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the steps of the multi-target tracking method based on compressed sensing and particle filtering of the present invention.

[0021] Another embodiment of the present invention provides a computer-readable storage medium item, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform steps of the multi-target tracking method based on compressed sensing and particle filtering of the present invention. Attached Figure Description

[0022] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating an embodiment of the multi-target tracking method based on compressed sensing and particle filtering provided by the present invention. Figure 2 This is an overall flowchart of another embodiment of the multi-target tracking system based on compressed sensing and particle filtering provided by the present invention; Figure 3 This is a comparison diagram of the target angle tracking effect of another embodiment of the multi-target tracking system based on compressed sensing and particle filtering provided by the present invention; Figure 4 This is a comparison chart of the target angular velocity tracking performance of another embodiment of the multi-target tracking method based on compressed sensing and particle filtering provided by the present invention; Figure 5 This is a comparison chart of the comprehensive performance indicators of target tracking for another embodiment of the multi-target tracking method based on compressed sensing and particle filtering provided by the present invention. Figure 6 This is a schematic diagram of another embodiment of the multi-target tracking system based on compressed sensing and particle filtering provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0026] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0028] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0029] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0030] See Figure 1 To address the multi-target tracking problem in existing technologies, an embodiment of the present invention provides a multi-target tracking method based on compressed sensing and particle filtering, comprising steps S1 to S4, the specific steps of which are as follows: S1. Compress the received raw array signal to obtain a low-dimensional compressed observation signal; S2. Initialize an independent particle swarm for each target to be tracked; each particle in the particle swarm contains the angle state variable and angular velocity state variable of the target to be tracked; S3. Based on the low-dimensional compressed observation signal, perform time recursive tracking on the particle swarm of each target to be tracked, and jointly calculate the weight of each particle in each particle swarm to obtain the weighted particle set of each target to be tracked. S4. Perform state estimation based on the weighted particle set of each target to be tracked, obtain the angle estimate and angular velocity estimate of each target to be tracked, and output the tracking result of each target to be tracked.

[0031] As another example of an embodiment of the present invention, see Figure 2 First, the particle swarm is initialized and partitioned according to the target. Each target to be tracked is assigned an independent particle swarm containing angle and angular velocity state variables. It is then determined whether the last frame of observation data has been processed. If not, the tracking process is executed in a time loop. Compressed observation data at that moment is obtained, and prediction is made based on the state at the previous moment. Strict physical constraints are imposed on the predicted angular velocity, and the weights of each particle are jointly calculated based on the current observation. The angle and angular velocity estimates of each target at the current moment are calculated using the complex averaging of angles and velocity deviation compensation. Intelligent resampling is performed based on the historical estimation error to adaptively adjust the process noise. After resampling, differentiated perturbations are injected for different targets to maintain the diversity of the particle swarm, and the state estimation results at that moment are output. Finally, after the time loop ends, the tracking results of all targets to be tracked are output.

[0032] In practical applications, two moving targets are simulated. Target 1 moves at a constant velocity in a straight line, with its azimuth angle changing from 20° to 60°. Target 2 is a maneuvering target, and its azimuth angle changes as follows: first stage (times 1-80) from 120° to 90° at a constant velocity; second stage (times 81-140) from 90° to 95° with slight maneuvering; and third stage (times 141-200) from 95° to 110° in the opposite direction. The initialization parameters are: number of array elements M=16, compression dimension P=8 (compression ratio 50%), number of targets K=2, number of particles per target N=100, and total tracking time T=200 sampling intervals. Normalized to 1, signal-to-noise ratio (SNR) = 10 dB, basic standard deviation of process noise. Adaptive adjustment coefficient Speed ​​penalty coefficient Regularization parameters Angular velocity constraint range [-5, 5]° / sampling; tracking estimation results are as follows Figure 3, Figure 4 As shown, compared with the estimation results of the traditional PF (Particle Filter) algorithm and the MUSIC (Multi-Signal Classification) + EKF (Extended Kalman Filter) algorithm, the present invention can maintain stable tracking of two targets throughout the tracking process and obtain smooth and accurate motion trajectories.

[0033] The tracking performance of this invention embodiment is quantitatively evaluated using root mean square error (RMSE) and computation time, and compared with two traditional algorithms, such as... Figure 5 As shown, the overall RMSE for angle tracking is 2.5240°, and the overall RMSE for angular velocity tracking is 0.4804° / sample, with a total computation time of 0.31 seconds. In comparison, the traditional PF algorithm has an angle tracking RMSE of 6.6467° and an angular velocity tracking RMSE of 1.9579° / sample, with a computation time of 0.53 seconds; the MUSIC+EKF algorithm has an angle tracking RMSE of 46.8527° and an angular velocity tracking RMSE of 3.7323° / sample, with a computation time of 0.38 seconds. The results show that the method of this invention achieves 50% data compression while improving angle tracking accuracy by 62.03%, angular velocity tracking accuracy by 75.46%, and computational efficiency by 42.67% compared to the traditional PF algorithm. Compared to the MUSIC+EKF algorithm... The method of this invention improves angle tracking accuracy by 94.61%, angular velocity tracking accuracy by 87.13%, and computational efficiency by 20.52%, fully verifying the comprehensive advantages of this invention in terms of accuracy, efficiency, and joint estimation capability. A phased analysis of the maneuvering characteristics of the target 2 trajectory shows that the patented method of this invention maintains low tracking errors at each stage and can maintain stable tracking even during target maneuvers and when approaching angular boundaries. In contrast, the traditional PF algorithm shows a significant increase in error during the target maneuvering stage, while the MUSIC+EKF algorithm exhibits large errors throughout the tracking process due to its two-stage processing structure and data association issues, especially near target maneuvers and angular boundaries. This further demonstrates the effectiveness of the complex averaging method, adaptive noise adjustment, and joint particle filter framework adopted in this invention.

[0034] In one embodiment, based on the low-dimensional compressed observation signal, time-recursive tracking is performed on the particle swarm of each target to be tracked, including steps S201 to S202, each step of which is as follows: S201. Based on the error between the historical state estimate and the prior value of each target to be tracked, dynamically adjust the initial process noise variance in the state prediction process to obtain the first process noise variance. Wherein, the first process noise variance It is dynamically adjusted based on recent estimation errors, and the calculation formula is as follows: in, Let be the estimation error of the k-th target at time t; W is the sliding window length. This is the state estimate of the k-th target at time t; The prior angular velocity; The basic noise variance; and These are the upper and lower limits for variance adjustment; For adjustment coefficients; The prior central value, combined with the formula, forms a closed-loop feedback; error Larger noise variance increases particle diffusion range to capture target maneuvers; error To reduce noise variance, particle aggregation is used to improve accuracy.

[0035] S202. Based on the first process noise variance, perform state prediction for each particle in the particle swarm of each target to be tracked, and apply physical range constraints to obtain the constrained predicted particle state.

[0036] For each particle i of each target k, state prediction is performed, calculated using the following formula: Physical constraints are imposed, and the predicted angular velocity is truncated to limit it to a physically feasible range. The calculation formula is as follows: in, ; This is process noise; The noise variance of the first process; and For the k-th target predicted at time t-1, the th The angular velocity and angular state of a particle at time t; and For the k-th target The angular velocity and angular state of each particle at time t-1; The sampling time interval; The applied physical constraints are used to prevent particles from spreading into unreasonable regions, accelerate convergence, and improve stability.

[0037] This invention introduces a closed-loop adaptive process noise adjustment mechanism in time-recursive tracking. Based on the error between historical state estimates and prior values, the process noise variance in the state prediction process is dynamically adjusted to obtain a first process noise variance that matches the current tracking state. The particle diffusion range is adaptively adjusted according to the actual motion error of the target, significantly enhancing the responsiveness to changes in the target's motion state and the robustness of tracking. Furthermore, based on the first process noise variance, state prediction is performed on each particle in the particle swarm, and a physically feasible range constraint is applied to the particle angular velocity after prediction. This effectively prevents the particle state from diffusing into unreasonable motion ranges, accelerates particle convergence to the true state, and significantly improves the rationality of state prediction and the stability of the tracking process.

[0038] In one embodiment, the weight of each particle in each particle swarm is jointly calculated to obtain a weighted particle set for each target to be tracked, including steps S301 to S303, each step of which is as follows: S301. Based on the predicted particle state, extract particles with the same index from the particle group of each target to be tracked to obtain the extracted particle group, and construct a joint guidance matrix based on the angular velocity state variables of each particle in the extracted particle group. The weight calculation is based on the likelihood function. For particle indexing Take the first particle from the particle swarm of each target. Each particle, its angle values ​​are combined into a vector. The joint guidance matrix is ​​constructed based on the combined vector angles, as shown in the following formula: in, For the time t, the first The joint guidance matrix of the joint particles; For the k-th target The angular state of a particle at time t; To correspond to the angle The guide vector; It is a complex matrix with M rows and K columns; S302. Estimate the signal amplitude based on the low-dimensional compressed observation signal, the joint steering matrix, and the preset compressed observation matrix; The compressed guidance matrix is ​​calculated using the following formula: The maximum a posteriori (MAP) or least squares (LS) estimate of the signal S(t) is obtained by the following formula: The solution is in, Let the angle state at time t be the first The compressed steering matrix of a joint particle; For the time t, the first The joint guidance matrix of the joint particles; It is a complex matrix with P rows and K columns; For the first The state assumptions of the joint particles are estimated; Y(t) is the actual compressed observation vector at time t; For theoretical observation models; This is the regularization matrix; S303. Calculate the observation residual based on the signal amplitude, calculate the joint basic weight of the extracted particle group by combining the predicted particle state with the signal energy penalty term, determine the corresponding particle weight of each target to be tracked based on the joint basic weight, and obtain the weighted particle set of each target to be tracked.

[0039] The residual after using the estimated signal is calculated using the following formula: Assuming the noise follows a Gaussian distribution, the likelihood function is: Meanwhile, to avoid numerical problems caused by excessively high estimated signal energy, an energy penalty term is added: The final formula for calculating the basic weights is as follows: in, For the first The observation residual vector of the joint particles; Y(t) is the actual compressed observation vector at time t; Let the angle state at time t be the first The compressed steering matrix of a joint particle; For the first State assumption estimation of a joint particle; For the first The state of a joint particle; It is the residual vector; For noise variance; This is the energy penalty coefficient; For the first The fundamental weights of the joint particles; This invention, through its embodiments, extracts particles with corresponding indices from independent particle swarms of each target based on the low-dimensional compressed observation signal, and constructs a joint steering matrix based on their angular states, thereby achieving accurate characterization of the observation signal under multi-target, low-dimensional observation conditions. Furthermore, based on the low-dimensional compressed observation signal, the joint steering matrix, and a preset compressed observation matrix, the signal amplitude at the prediction time is estimated, embedding the compressed sensing reconstruction process into particle weight evaluation, significantly improving the estimation accuracy of signal amplitude under dimensionality reduction observation. The observation residual is calculated based on the signal amplitude, and the basic weight of the particles is calculated in conjunction with the signal energy penalty term to obtain weighted particles. This allows the residual to reflect the matching degree between the particle state and the actual observation, and the energy penalty term to suppress unreasonable estimations, thus achieving more accurate likelihood state estimation under low-dimensional observation conditions.

[0040] In one embodiment, determining the weight of the corresponding particle in each target particle swarm to be tracked based on the joint basic weights further includes steps S401 to S402, each step being as follows: S401. Calculate the velocity penalty factor using the velocity penalty term formula based on the predicted particle state. The formula for the speed penalty term is as follows: in, This is the speed penalty coefficient. The range of prior angular velocities; For the k-th target to be tracked in the predicted particle state The angular velocity state variable of a particle at time t; K is the number of targets to be tracked; This is a speed penalty factor.

[0041] S402. The joint base weights are corrected by the velocity penalty factor to determine the corresponding particle weights for each target to be tracked.

[0042] The final weight calculation is performed using the following formula: The normalization calculation is performed using the following formula: in, For the first Normalized weights of the joint particles; For the first The fundamental weights of the joint particles; As a speed penalty factor; In one embodiment, state estimation is performed based on the weighted particle set of each target to be tracked to obtain the angle estimate and angular velocity estimate of each target to be tracked, and the tracking result of each target to be tracked is output, including steps S501 to S503, each step of which is as follows: S501. Based on the weighted particles, the complex average method is used to map the values ​​of the angle state variables in the weighted particles onto a complex unit circle for weighted averaging to obtain a complex result. The complex result is then converted back into an angle value to obtain an estimated angle value for the target to be tracked. The formula for calculating the angle estimate is as follows: in, The intermediate complex variable is used for estimating the angle of the k-th target. For the first Normalized weights of the joint particles; For the k-th target The angular state of each particle at time t; This is a complex number that maps angle values ​​from degrees to radians onto the unit circle. This is the final angle estimate of the k-th target at time t; To take complex numbers Argument; S502. Based on the weighted particles, the values ​​of the angular velocity state variables in the weighted particles are weighted and averaged to obtain a weighted average result. The weighted average result is then fine-tuned through a compensation mechanism to obtain an estimated value of the angular velocity of the target to be tracked. The angular velocity estimation uses a weighted average with bias compensation, and the calculation formula is as follows: The current estimate is fine-tuned based on recent estimation bias to further suppress error accumulation. The calculation formula is as follows: The estimated angular velocity is obtained after compensation, and the calculation formula is as follows: ,in For the compensation coefficient ( ) in, For the k-th target The angular velocity of a particle at time t; This is the state estimate of the k-th target at time t; is the recent average deviation of the estimated angular velocity of the k-th target; W is the sliding window length; The prior angular velocity; This is the estimated final angular velocity of the k-th target at time t; S503. Based on the estimated angle and the estimated angular velocity, output the tracking result for each target to be tracked.

[0043] The output is the state estimate at the current time t: These estimates will be transmitted to a display terminal or host computer system for real-time display or further processing.

[0044] This invention employs a complex averaging method for angle state estimation. Based on the weighted particles, the values ​​of the angle state variables of each particle are mapped onto a complex unit circle and weighted averaged. The resulting complex value is then converted back into an angle value, yielding an estimated angle of the target to be tracked. This method addresses the periodicity of angle variables and effectively eliminates estimation ambiguity at the 360° boundary caused by direct linear averaging, ensuring the accuracy of angle estimation. Furthermore, based on the weighted particles, the values ​​of the angular velocity state variables of each particle are weighted averaged to obtain a preliminary result. This weighted average result is then fine-tuned based on a historical deviation compensation mechanism to obtain an estimated angular velocity of the target to be tracked. This effectively suppresses error accumulation and deviation caused by model approximation or noise during long-term tracking, significantly improving the accuracy of angular velocity estimation and the overall robustness of the tracking process.

[0045] In one embodiment, after obtaining the weighted particle set for each target to be tracked, intelligent resampling is further included, comprising steps S601 to S602, each step being as follows: S601. Based on the weighted particle set of each target to be tracked, a new particle set for each target to be tracked is generated using a system resampling method; S602. Randomly select a portion of the new particles in the new particle set, and apply differential perturbations to the state variables of the new particles to maintain particle diversity.

[0046] Among them, according to the normalized weights Perform system resampling to generate a new particle set index. For the resampled particles, apply differential perturbations according to the target; for target k, randomly select... The particles are perturbed: ,in Using the process noise variance at time t Among them, the disturbance factor The perturbation factor can be set for different targets; that is, it is larger when estimating targets with high uncertainty.

[0047] By performing intelligent resampling and differential perturbation before outputting the tracking results, the embodiments of the present invention can more effectively maintain the uniformity and representativeness of the particle set, thereby alleviating the particle degradation problem and providing a more reliable state sample basis for subsequent tracking steps.

[0048] In one embodiment, the received raw array signal is compressed and observed to obtain a low-dimensional compressed observation signal, including steps S701 to S702, each of which is as follows: S701. Construct an original received signal vector based on the received original array signal; the original received signal vector represents the superposition of signals and noise from multiple targets to be tracked; In this model, a full-dimensional array signal model is established. Assume a uniform linear array (ULA) with M elements, receiving K far-field narrowband signals. The array reception vector at time t is: in, This is the array received signal vector; It is the angle vector of K targets; It is the guidance matrix, and its k-th column is the guidance vector. (To simplify the formula, it is often assumed that...) ,but ; It is the complex envelope vector of the signal source; It is an additive white Gaussian noise (AWGN) vector, and its covariance matrix is... ; S702. By using a preset compressed observation matrix, the original received signal vector is linearly reduced in dimension and projected to obtain a low-dimensional compressed observation signal; wherein, the low-dimensional compressed observation signal includes a compressed observation signal vector; the number of rows of the compressed observation signal vector is much smaller than the number of array elements.

[0049] This involves establishing a compressed observation model and introducing a compressed observation matrix. The original signal is linearly reduced in dimension by projection to obtain a compressed observation vector. : like Designed to meet (If a random Gaussian matrix is ​​used and normalized), then The covariance is still The compressed observation model can be simplified to: in, By reducing the data dimension from M to P, the amount of data processed in the backend is significantly reduced. The particle state is initialized based on the low-dimensional compressed observation signal, and a state-space model is defined: The state equation is For the k-th objective, its state is defined as follows: (Angle and angular velocity), taking the uniform velocity (CV) model as an example: in, The sampling interval (usually normalized to 1). It has a mean of zero and a variance of . Process noise, ; Based on the compressed observation model ,in Perform particle filter initialization for each target. Based on prior information (such as the initial angle) and initial angular velocity Initialize its particle swarm: in, , where represents the initial angle and initial angular velocity sampled from the Gaussian distribution; N is the number of particles for each target; the initialized particle swarm will enter a time-recursive tracking process.

[0050] This invention constructs an original received signal vector containing multiple target signals and noise superposition, and then uses a preset compressed observation matrix to perform linear dimensionality reduction projection on the vector to obtain a low-dimensional compressed observation signal with a vector row count much smaller than the number of array elements. This achieves effective compression of the original array signal and significantly reduces the burden of data acquisition and transmission.

[0051] like Figure 6 Based on the multi-target tracking method using compressed sensing and particle filtering, a corresponding system implementation is provided. This invention provides a multi-target tracking system based on compressed sensing and particle filtering, including: a compressed observation module 801, an initialization particle module 802, a weight calculation module 803, and a state estimation module 804; The compressed observation module 801 is used to perform compressed observation on the received original array signal to obtain a low-dimensional compressed observation signal. The initialization particle module 802 is used to initialize an independent particle swarm for each target to be tracked; each particle in the particle swarm contains the angle state variable and angular velocity state variable of the target to be tracked. The weight calculation module 803 is used to perform time recursive tracking of the particle swarm of each target to be tracked according to the low-dimensional compressed observation signal, and jointly calculate the weight of each particle in each particle swarm to obtain a weighted particle set of each target to be tracked. The state estimation module 804 is used to perform state estimation based on the weighted particle set of each target to be tracked, obtain the angle estimate and angular velocity estimate of each target to be tracked, and output the tracking result of each target to be tracked.

[0052] This invention provides a low-dimensional compressed observation signal by compressing the received raw array signal. This significantly reduces the data dimensionality and transmission burden by drastically compressing the sensor data volume. A distributed state representation framework for multiple targets is established by initializing an independent particle swarm for each target, with each particle containing the target's angle and angular velocity state variables. This reduces computational complexity from exponential to linear and provides a state carrier for subsequent joint angle and velocity estimation. Based on the low-dimensional compressed observation signal, time-recursive tracking is performed on the particle swarm, and a joint processing mechanism is used in the weight calculation stage to significantly improve the accuracy of likelihood state estimation. The state estimation yields the angle and angular velocity estimates for each target, and the tracking results are output, achieving high-precision joint tracking of the angles and velocities of multiple moving targets with lower data dimensionality and computational resources.

[0053] It is understood that the above system embodiments correspond to the method embodiments of the present invention, and can implement the multi-target tracking method based on compressed sensing and particle filtering provided by any of the above method embodiments of the present invention.

[0054] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0055] For ease of description and brevity, the system embodiments of the present invention include all the implementation methods described above in the embodiments of the multi-target tracking method based on compressed sensing and particle filtering, and will not be repeated here.

[0056] Based on the above embodiments of the multi-target tracking method based on compressed sensing and particle filtering, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the multi-target tracking method based on compressed sensing and particle filtering according to any embodiment of the present invention.

[0057] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0058] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0059] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0060] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the multi-target tracking method based on compressed sensing and particle filtering described in any of the above-described method embodiments of the present invention.

[0061] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0062] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A multi-target tracking method based on compressed sensing and particle filtering, characterized in that, include: Compressed observations are performed on the received raw array signals to obtain low-dimensional compressed observation signals; A separate set of particle swarms is initialized for each target to be tracked; each particle in the particle swarm contains the angle state variable and angular velocity state variable of the target to be tracked; Based on the low-dimensional compressed observation signal, time recursive tracking is performed on the particle swarm of each target to be tracked, and the weight of each particle in each particle swarm is jointly calculated to obtain the weighted particle set of each target to be tracked. State estimation is performed based on the weighted particle set of each target to be tracked, and the angle estimate and angular velocity estimate of each target to be tracked are obtained. The tracking result of each target to be tracked is then output.

2. The multi-target tracking method based on compressed sensing and particle filtering as described in claim 1, characterized in that, The step of performing time-recursive tracking of the particle swarm for each target to be tracked based on the low-dimensional compressed observation signal includes: The initial process noise variance in the state prediction process is dynamically adjusted based on the error between the historical state estimate and the prior value of each target to be tracked, to obtain the first process noise variance; Based on the first process noise variance, state prediction is performed for each particle in the particle swarm of each target to be tracked, and physical range constraints are applied to obtain the constrained predicted particle state.

3. The multi-target tracking method based on compressed sensing and particle filtering as described in claim 2, characterized in that, The joint calculation of the weight of each particle in each particle swarm to obtain a weighted particle set for each target to be tracked is specifically as follows: Based on the predicted particle state, particles with the same index are extracted from the particle swarm of each target to be tracked to obtain an extracted particle group. A joint guidance matrix is ​​constructed based on the angular velocity state variables of each particle in the extracted particle group. The signal amplitude is estimated based on the low-dimensional compressed observation signal, the joint steering matrix, and the preset compressed observation matrix; The observation residual is calculated based on the signal amplitude. The joint basic weight of the extracted particle group is calculated by combining the predicted particle state with the signal energy penalty term. Based on the joint basic weight, the corresponding particle weight of each target to be tracked is determined, and the weighted particle set of each target to be tracked is obtained.

4. The multi-target tracking method based on compressed sensing and particle filtering as described in claim 3, characterized in that, The step of determining the weight of each particle in the target particle swarm to be tracked based on the joint basic weights further includes: Based on the predicted particle state, the velocity penalty factor is calculated using the velocity penalty term formula; The joint base weights are corrected by the velocity penalty factor to determine the corresponding particle weights for each target to be tracked. The formula for the speed penalty term is as follows: in, This is the speed penalty coefficient. The range of prior angular velocities; Let be the angular velocity state variable of the i-th particle of the k-th target to be tracked in the predicted particle state at time t; K is the number of targets to be tracked; This is a speed penalty factor.

5. The multi-target tracking method based on compressed sensing and particle filtering as described in claim 1, characterized in that, The process involves performing state estimation based on the weighted particle set of each target to obtain the angle and angular velocity estimates for each target, and outputting the tracking result for each target. Specifically: Based on the weighted particles, the complex average method is used to map the values ​​of the angle state variables in the weighted particles onto a complex unit circle for weighted averaging to obtain a complex result. The complex result is then converted back into an angle value to obtain an estimated angle value for the target to be tracked. Based on the weighted particles, the values ​​of the angular velocity state variables in the weighted particles are weighted and averaged to obtain a weighted average result. The weighted average result is then fine-tuned through a compensation mechanism to obtain an estimated value of the angular velocity of the target to be tracked. Based on the estimated angle and the estimated angular velocity, the tracking result for each target to be tracked is output.

6. The multi-target tracking method based on compressed sensing and particle filtering as described in claim 1, characterized in that, After obtaining the weighted particle set for each target to be tracked, intelligent resampling is also included: Based on the weighted particle set of each target to be tracked, a new particle set for each target to be tracked is generated using a system resampling method; Randomly select a portion of new particles from the new particle set, and apply differential perturbations to the state variables of the new particles to maintain particle diversity.

7. The multi-target tracking method based on compressed sensing and particle filtering as described in claim 1, characterized in that, The process of compressing the received raw array signal to obtain a low-dimensional compressed observation signal specifically involves: Based on the received original array signals, an original received signal vector is constructed; the original received signal vector represents the superposition of signals and noise from multiple targets to be tracked. By using a preset compressed observation matrix, the original received signal vector is linearly reduced in dimension and projected to obtain a low-dimensional compressed observation signal; wherein, the low-dimensional compressed observation signal includes a compressed observation signal vector; the number of rows of the compressed observation signal vector is much smaller than the number of array elements.

8. A multi-target tracking system based on compressed sensing and particle filtering, characterized in that, include: The module includes a compressed observation module, a particle initialization module, a weight calculation module, and a state estimation module. The compressed observation module is used to perform compressed observation on the received original array signal to obtain a low-dimensional compressed observation signal. The initialization particle module is used to initialize an independent particle swarm for each target to be tracked; each particle in the particle swarm contains the angle state variable and angular velocity state variable of the target to be tracked. The weight calculation module is used to perform time recursive tracking of the particle swarm of each target to be tracked based on the low-dimensional compressed observation signal, and jointly calculate the weight of each particle in each particle swarm to obtain a weighted particle set of each target to be tracked. The state estimation module is used to perform state estimation based on the weighted particle set of each target to be tracked, obtain the angle estimate and angular velocity estimate of each target to be tracked, and output the tracking result of each target to be tracked.

9. A terminal device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the multi-target tracking method based on compressed sensing and particle filtering as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the multi-target tracking method based on compressed sensing and particle filtering as described in any one of claims 1-7.

Images