Distributed multi-target fusion tracking method and device based on time calibration
By employing time calibration and data fusion techniques, the problem of low data fusion efficiency in distributed multi-target tracking was solved, achieving efficient and accurate multi-target state estimation and improving target positioning accuracy and computational efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2025-12-29
- Publication Date
- 2026-05-08
AI Technical Summary
In distributed multi-target tracking scenarios, existing technologies struggle to efficiently and accurately fuse data from multiple sensor nodes, leading to disruption of target trajectory continuity, target mistracking or loss, and low computational efficiency, making it difficult to meet real-time requirements.
A time-calibrated distributed multi-target fusion tracking method is adopted. The method utilizes the generalized covariance cross-fusion criterion and Gaussian mixture approximation technique to construct the posterior probability density expression of the time calibration parameters. Time calibration is performed by selecting the mean of the Gaussian component with the largest weight, and data fusion is performed based on the generalized covariance cross-fusion criterion to achieve multi-target state estimation.
It enables efficient and accurate fusion of data from multiple sensor nodes in distributed multi-target tracking scenarios, improving target localization accuracy, reducing computational costs, and meeting real-time requirements.
Smart Images
Figure CN121999009A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of sensor technology, and in particular to a time-calibrated distributed multi-target fusion tracking method and apparatus. Background Technology
[0002] Distributed multi-target tracking technology plays a crucial role in fields such as military reconnaissance, intelligent transportation, and drone collaboration. It achieves stable tracking of dynamic targets by fusing data from multiple sensor nodes. However, with the expansion of sensor network scale and the increase in heterogeneity, the tracking system faces multiple challenges.
[0003] The asynchronous nature of time leads to inconsistencies in sampling periods, start times, and transmission / processing delays among sensor nodes, making it difficult to align data in the time dimension (e.g., different sampling frequencies between radar and camera, or out-of-order wireless transmission), thus disrupting the continuity of target trajectories. Existing centralized synchronization methods rely on a global clock or frequent calibration, resulting in high computational and communication costs and difficulty in meeting real-time requirements. Data association ambiguity stems from significant differences in the observation characteristics of the same target by heterogeneous sensors (radar, infrared, vision), coupled with noise interference, making it difficult to match cross-modal data (e.g., point clouds and images). In related technologies, target mistracking or loss often occurs due to mismatched feature dimensions or fuzzy association rules. The computational efficiency bottleneck arises because centralized architectures require transmitting data from all nodes to a central processor, with the data volume increasing exponentially with the number of nodes, leading to processing delays. Furthermore, traditional algorithms are extremely complex in multi-target scenarios, making them unsuitable for the real-time demands of highly dynamic environments.
[0004] Therefore, in distributed multi-target tracking scenarios, how to efficiently and accurately fuse data from multiple sensor nodes to achieve precise positioning of the tracked target is a technical problem that urgently needs to be solved. Summary of the Invention
[0005] This invention provides a time-calibrated distributed multi-target fusion tracking method and apparatus to address the aforementioned deficiencies of the prior art. In distributed multi-target tracking scenarios, it efficiently and accurately fuses data from multiple sensor nodes to achieve precise positioning of the tracked target.
[0006] This invention provides a time-calibrated distributed multi-target fusion tracking method, comprising the following steps.
[0007] Based on the generalized covariance cross-fusion criterion, a posterior probability density expression for the time calibration parameter is constructed using Gaussian mixture approximation and probability hypothesis density filters. The posterior probability density expression is in the form of a Gaussian mixture exponential function. The time calibration parameter is used to align the first multi-target probability density function and the second multi-target probability density function in the time dimension. The first multi-target probability density function is obtained by filtering data collected from a reference sensor, and the second multi-target probability density function is obtained by filtering data collected from the sensor to be registered. The mean of the Gaussian component with the largest weight is selected from the multiple Gaussian components of the posterior probability density expression as the estimated value of the time calibration parameter. Based on the estimated value of the time calibration parameter, the second multi-target probability density function is aligned with the first multi-target probability density function in time to obtain the calibrated second multi-target probability density function. Based on the generalized covariance cross-fusion criterion, the first multi-target probability density function and the calibrated second multi-target probability density function are fused to obtain the multi-target state estimate of the tracked target.
[0008] According to the distributed multi-target fusion tracking method based on time calibration provided by the present invention, the step of constructing a posterior probability density expression for the time calibration parameters based on the generalized covariance cross-fusion criterion and utilizing Gaussian mixture approximation techniques and probability hypothesis density filters includes: The initial posterior probability density expression of the time calibration parameter is determined using the generalized covariance cross-fusion criterion. This initial posterior probability density expression includes the prior probability density function of the time calibration parameter and the likelihood function between the first and second multi-objective probability density functions with respect to the time calibration parameter. The Gaussian mixture approximation technique and probability hypothesis density filter are used to determine the Gaussian mixture exponential form of the likelihood function. Based on the likelihood function in the Gaussian mixture exponential form and the prior probability density function, the posterior probability density expression of the time calibration parameter in the Gaussian mixture exponential form is obtained.
[0009] According to the time-calibrated distributed multi-target fusion tracking method provided by the present invention, the step of determining the Gaussian mixture exponential form of the likelihood function using Gaussian mixture approximation technique and probability hypothesis density filter includes: determining the mathematical expression of the likelihood function using probability hypothesis density filter; and expressing the mathematical expression of the likelihood function as a Gaussian mixture exponential form using Gaussian mixture approximation technique.
[0010] According to the time-calibrated distributed multi-target fusion tracking method provided by the present invention, the step of determining the mathematical expression of the likelihood function using a probability hypothesis density filter includes: After processing the first multi-objective probability density function with a probability hypothesis density filter, a first intensity function is obtained; after processing the second multi-objective probability density function with a probability hypothesis density filter, a second intensity function is obtained; the first intensity function and the second intensity function are used to represent the likelihood function, and the mathematical expression of the likelihood function is obtained.
[0011] According to the time-calibrated distributed multi-target fusion tracking method provided by the present invention, the likelihood function is expressed by the following formula: ; in, ; in, For the first intensity function, This is the second intensity function under the time calibration parameters. To track the state variables of the target; These are variables related to time calibration parameters; The weighting coefficients are used; the Gaussian mixture exponential form of the likelihood function is expressed by the following formula: ; Where a and b are the index numbers, The number of Gaussian components corresponding to the first multi-objective probability density function. The number of Gaussian components corresponding to the second multi-objective probability density function. These are weighting coefficients used to balance the contributions of different terms to the objective function. It follows a normal distribution. For time calibration parameters, The mean of the Gaussian components is given. Let V be the variance of the Gaussian components.
[0012] According to the time-calibrated distributed multi-target fusion tracking method provided by the present invention, the step of fusing the first multi-target probability density function and the calibrated second probability hypothesis density filter multi-target probability density function based on the generalized covariance cross-fusion criterion to obtain the multi-target state estimate of the tracked target includes: Based on the generalized covariance cross-fusion criterion, the first multi-target probability density function and the calibrated second multi-target probability density function are fused to obtain the Gaussian components of the multi-target state association of the tracked target; from the Gaussian components of the multi-target state association, the mean of the Gaussian components with a weight value greater than a preset threshold is selected as the multi-target state estimate of the tracked target.
[0013] The present invention also provides a time-calibrated distributed multi-target fusion tracking device, comprising: The system comprises the following modules: a construction module for constructing a posterior probability density expression for a time calibration parameter based on a generalized covariance cross-fusion criterion, using Gaussian mixture approximation techniques and a probability hypothesis density filter; wherein the posterior probability density expression is in the form of a Gaussian mixture exponential function; the time calibration parameter is used to align a first multi-target probability density function and a second multi-target probability density function in the time dimension; the first multi-target probability density function is obtained by filtering data collected by a reference sensor, and the second multi-target probability density function is obtained by filtering data collected by the sensor to be registered; a selection module for selecting the mean of the Gaussian component with the largest weight from multiple Gaussian components of the posterior probability density expression as the estimated value of the time calibration parameter; a synchronization module for aligning the second multi-target probability density function with the first multi-target probability density function in time based on the estimated value of the time calibration parameter, thus obtaining a calibrated second multi-target probability density function; and a fusion module for fusing the first multi-target probability density function and the calibrated second multi-target probability density function based on the generalized covariance cross-fusion criterion to obtain a multi-target state estimate of the tracked target.
[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the time-calibrated distributed multi-target fusion tracking method as described above.
[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the time-calibrated distributed multi-target fusion tracking method as described above.
[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the time-calibrated distributed multi-target fusion tracking method as described above.
[0017] This invention provides a distributed multi-target fusion tracking method and apparatus based on time calibration. It utilizes the Generalized Covariance Cross (GCI) fusion criterion and Gaussian mixture approximation to construct the posterior probability density expression of time calibration parameters. The Gaussian mixture approximation can represent a complex probability distribution as a weighted sum of multiple Gaussian components, providing a more detailed description of the probability distribution of the time calibration parameters and enabling faster convergence in the time calibration parameter estimation process. This allows for a more comprehensive and accurate consideration of the correlation and uncertainty between data from different sensors. The mean of the Gaussian component with the largest weight value is selected from the posterior probability density expression as the estimated value of the time calibration parameter. Based on the probability maximization principle, the most probable value is effectively selected from multiple possible calibration parameter values, improving the accuracy and reliability of the time calibration parameter estimation. Based on the estimated value of the time calibration parameter, the first and second multi-target probability density functions are synchronized to ensure that the data from different sensors are consistent in the time dimension, enabling subsequent fusion operations to be performed under a unified time reference. Accurate time synchronization is the foundation of multi-sensor data fusion, avoiding target state estimation deviations caused by time asynchrony. Based on the Generalized Covariance Cross-Fusion Criterion (GCI), the first multi-target probability density function and the calibrated second multi-target probability density function are fused. The GCI criterion fully considers the uncertainties of each probability hypothesis density filter during the fusion process, generating more accurate and comprehensive multi-target state estimation results by reasonably allocating weights and adjusting covariance. Therefore, this embodiment of the invention can achieve high-precision joint estimation of time calibration parameters and multi-target states with relatively low computational cost, efficiently and accurately fusing data from multiple sensor nodes in distributed multi-target tracking scenarios to achieve precise positioning of the tracked target. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0019] Figure 1 This is a flowchart illustrating the distributed multi-target fusion tracking method based on time calibration provided by the present invention.
[0020] Figure 2 This is a flowchart illustrating the method for constructing the posterior probability density expression of time calibration parameters using Gaussian mixture approximation, as provided by the present invention.
[0021] Figure 3 This is a schematic diagram of the distributed multi-target fusion tracking process based on time calibration provided by the present invention.
[0022] Figure 4 It is a schematic diagram of the trajectory of the target in a two-dimensional plane.
[0023] Figure 5 This is a schematic diagram illustrating the estimation of time calibration parameters obtained using embodiments of the present invention.
[0024] Figure 6 This is a comparison chart of target OSPA errors in a two-dimensional plane between the embodiments provided by this invention and the multi-target fusion methods implemented by various related technologies.
[0025] Figure 7 This is a comparison chart of the target GOSPA error in a two-dimensional plane between the embodiments provided by this invention and the multi-target fusion method implemented by various related technologies.
[0026] Figure 8 This is a schematic diagram of the structure of the distributed multi-target fusion tracking device based on time calibration provided by the present invention.
[0027] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0029] The following is combined Figures 1-7 This invention describes a time-calibrated distributed multi-target fusion tracking method.
[0030] Figure 1 This is a flowchart illustrating the distributed multi-target fusion tracking method based on time calibration provided by the present invention.
[0031] Step 101: Based on the generalized covariance cross-fusion criterion, construct the posterior probability density expression of the time calibration parameter using Gaussian mixture approximation technique and probability hypothesis density filter.
[0032] The posterior probability density expression is in the form of a Gaussian mixture exponential function; the time calibration parameter is used to align the first multi-objective probability density function and the second multi-objective probability density function in the time dimension; the first multi-objective probability density function is obtained by filtering the data collected by the reference sensor, and the second multi-objective probability density function is obtained by filtering the data collected by the sensor to be registered.
[0033] The generalized covariance cross-fusion criterion is a distributed data fusion method used to merge observation data from different sensors in a multi-sensor system, while considering the correlation between sensors. Its core idea is to construct a joint covariance constraint with a consistent upper bound, enabling robust fusion of estimates from different sensors with minimal mean square error without underestimating uncertainty.
[0034] Time calibration parameters are used to characterize the time offset between the reference sensor and the sensor to be registered, and are used to eliminate time asynchrony problems caused by differences in sampling period, transmission delay or system processing time.
[0035] The posterior probability density expression for the time calibration parameter describes the conditional probability distribution of the time calibration parameter under observed data, representing the estimation uncertainty of the time bias.
[0036] A reference sensor, which serves as a time base, has its observation data timestamps considered as standard time (base time).
[0037] The sensor to be registered, the sensor whose time offset needs to be calibrated, and its observation data need to be aligned with the reference sensor.
[0038] The sensor to be registered and the reference sensor jointly acquire the state data (such as position and velocity) of the target being tracked, but there is a time asynchrony problem.
[0039] The target being tracked is a dynamic object (such as a vehicle, drone, or pedestrian) that is being tracked collaboratively by multiple sensors. Figure 4 A schematic diagram of the track of the target in a two-dimensional plane is shown.
[0040] In the embodiments provided by this invention, the posterior probability density expression is in the form of a Gaussian mixture exponential function. For an embodiment regarding the construction of the posterior probability density expression for the time calibration parameters, see [link to embodiment]. Figure 2 The relevant content will not be repeated here.
[0041] Step 102: Select the mean of the Gaussian component with the largest weight from the multiple Gaussian components of the posterior probability density expression as the estimated value of the time calibration parameter.
[0042] Step 103: Based on the estimated value of the time calibration parameter, align the second multi-objective probability density function with the first multi-objective probability density function in time to obtain the calibrated second multi-objective probability density function.
[0043] This step is represented by the following formula: Among them, among them, Let be the observed data vector, representing the point in the state space of interest. For time calibration parameters, For time calibration parameters δ ( θ ˉ) Related prior probabilities, These are the mixing coefficients, used to weight the individual Gaussian components, based on a multivariate Gaussian distribution. Representing observation data x The mean vector after calibration parameters are adjusted at a given time. Covariance Matrix The probability density is given below.
[0044] in, ; ; in, This is the original mean vector. A This is the transformation matrix used for time calibration parameters. Perform a linear transformation, and then adjust the mean vector using an exponential mapping. The adjusted mean is obtained. , It is the original covariance matrix. A Similarly, for matrices used in linear transformations, the original covariance matrix is transformed through an exponential mapping. It is a time calibration parameter The associated noise covariance matrix is used to represent additional noise or uncertainty.
[0045] Step 104: Based on the generalized covariance cross-fusion criterion, fuse the first multi-target probability density function and the calibrated second multi-target probability density function to obtain the multi-target state estimate of the tracked target.
[0046] In some embodiments, the first multi-target probability density function and the calibrated second multi-target probability density function can be fused based on the generalized covariance cross-fusion criterion to obtain the Gaussian components of the multi-target state association of the tracked target; from the Gaussian components of the multi-target state association, the mean of the Gaussian components with a weight value greater than a preset threshold is selected as the multi-target state estimate of the tracked target.
[0047] Based on the generalized covariance cross-fusion criterion, the first multi-objective probability density function and the calibrated second multi-objective probability density function are fused to obtain the fused probability hypothesis density function, which is obtained by weighted summation of multiple Gaussian components. Each component corresponds to the state distribution of a potential target, as expressed by the following formula: ; Where x: the observation data vector, representing a point in the state space of interest; and : These represent the number of Gaussian components corresponding to the first and second multi-objective probability density functions, respectively; The weighting coefficient is the time calibration parameter δ( The function is used to adjust the contribution of each Gaussian component in the fusion result; It is a multivariate Gaussian distribution, meaning that the mean is... Covariance Matrix The probability density of the observed data x.
[0048] in, The calculation formula is as follows: ; and : These are the original mixing coefficients of the a-th and b-th Gaussian components in the two information sources (the first multi-objective probability density function and the second multi-objective probability density function); ω is a weighting parameter used to balance the contributions of the two information sources; The prior probabilities related to the time calibration parameters; This is the kernel function, used to adjust the calculation of the weighting coefficients; and , are the covariance matrices of the a-th and b-th Gaussian components in the two information sources, respectively. It is the covariance matrix after adjustment by time calibration parameters; and These are the mean vectors of the a-th and b-th Gaussian components from the two information sources, respectively. It is the mean vector adjusted by the time calibration parameter. Where: ; ; in, ω is a weighting parameter used to balance the covariance information of the two information sources.
[0049] After obtaining the Gaussian components corresponding to each target state of the tracked target using the above formula, the multi-target state estimation of the tracked target can be determined based on the Gaussian components corresponding to each target state of the tracked target.
[0050] In the specific implementation process, the mean of the Gaussian components with weight values greater than a preset threshold (e.g., 0.5) can be selected from the Gaussian components associated with the multi-target state as the estimated value of the multi-target state of the tracking target.
[0051] In specific implementation, the embodiments provided by this invention can be derived from... Figure 3 The three modules shown are executed as follows: Module 1 performs local filtering. This module receives raw sensor data and applies various filtering algorithms (such as Kalman filter (KF), probability hypothesis density filter (PHD), and potential balance probability hypothesis density filter (CPHD)) to process the data, obtaining multi-target probability density data. These filtering algorithms aim to remove noise, improve data quality, and provide more accurate basic data for subsequent processing.
[0052] The multi-target probability density data obtained after filtering is passed to module 2 for further processing.
[0053] Module 2 receives multi-target probability density data from Module 1, performs time bias estimation operation (steps 101-102), and determines the estimated value of the time calibration parameter.
[0054] Module 3 receives the estimated value of the time calibration parameters from Module 2 and executes steps 103-104. Based on the GCI criterion, it fuses the time-calibrated multi-target probability density data to achieve accurate state estimation of the tracked target.
[0055] In addition, the fused data output by module 3 will be fed back to module 1 so that module 1 can use the fused data as a reference when performing subsequent filtering operations.
[0056] like Figure 5 As shown, in the process of multiple iterative calculations (estimation of time calibration parameters for multiple acquisitions of sensor data), the estimated values of time calibration parameters obtained by the embodiments of the present invention have the following advantages: It exhibits good convergence characteristics: the four curves in the figure (red and purple represent estimated values, blue and black represent true values) all show a trend of gradually stabilizing with the increase of the number of iterations. This indicates that during the iteration process, the estimated value of the time calibration parameter continuously approaches the true value, and the algorithm used (such as the fusion algorithm based on Gaussian mixture approximation and GCI criterion) can effectively estimate the time calibration parameter and has convergence.
[0057] It has a fast convergence speed: the estimated curves in different subgraphs begin to approach the true curves with fewer iterations, indicating that the algorithm converges quickly.
[0058] High accuracy: The estimated curves in each subplot deviate little from the true curves, indicating that the embodiments provided by this invention can accurately estimate time calibration parameters.
[0059] Figure 6The illustration shows a comparison of target OSPA errors in a two-dimensional plane between embodiments provided by this invention and multi-target fusion methods implemented using various related technologies. For example... Figure 6 As shown: Multi-target fusion method based on single-node filtering: The blue dashed line in the figure represents the OSPA error of single-node filtering. It can be seen that the overall error value is relatively high and fluctuates significantly. This indicates that relying solely on filtering from a single sensor has significant limitations in target tracking and state estimation, failing to fully utilize multi-sensor information, resulting in inaccurate and unstable estimation of the target state.
[0060] Multi-target fusion method without considering time calibration parameter estimation: The pink dashed line represents the case where time calibration parameter estimation is not considered. The error is also quite significant in certain time periods, indicating that without estimation and adjustment of time calibration parameters, different sensors, due to time asynchrony and other factors, will cause deviations in the estimated target state after fusion, affecting tracking accuracy.
[0061] The multi-target fusion method provided by this invention: The red dashed line represents the case where time calibration parameter estimation is considered. Compared with the previous two methods, its OSPA error is reduced to a certain extent overall, especially at some key time points, where the peak error is significantly reduced. This indicates that by estimating the time calibration parameter and calibrating the sensor data, the accuracy of multi-sensor data fusion can be effectively improved, making the target state estimation closer to the true value.
[0062] Multi-target fusion method with known time calibration parameters: The green solid line represents the ideal case where the time calibration parameters are known. Its error is relatively low and relatively stable, demonstrating that under ideal time synchronization conditions, multi-sensor data fusion can achieve good target tracking results. It also shows that methods considering time calibration parameter estimation are approaching the ideal situation, verifying the effectiveness and necessity of time calibration parameter estimation.
[0063] Figure 7 This diagram illustrates a comparison of target GOSPA errors in a two-dimensional plane between embodiments provided by this invention and multi-target fusion methods implemented using various related technologies. Four sub-figures respectively demonstrate the changes of different types of errors (such as GOSPA error, position error, missed detection error, false alarm error, etc.) over time. Figure 7 As shown: A multi-target fusion method based on single-node filtering: The blue dashed line in the figure represents the error of single-node filtering. It can be seen that in each sub-figure, the error value of single-node filtering is generally large and fluctuates drastically. This indicates that relying solely on a single sensor for filtering, due to the lack of fusion and coordination of multi-sensor information, makes it difficult to accurately estimate the target state, resulting in large errors.
[0064] Multi-target fusion method without considering time calibration parameter estimation: The pink dashed line represents the case without considering time calibration parameter estimation. Its error is significantly higher than the case considering time calibration parameter estimation in certain time periods, indicating that factors such as time asynchrony have a significant impact on the accuracy of multi-sensor data fusion. Without estimating and adjusting the time calibration parameters, data from different sensors will deviate during fusion, thus reducing the accuracy of target tracking.
[0065] The multi-target fusion method provided in this embodiment of the invention: the red dashed line represents the case considering the estimation of time calibration parameters. Compared with the previous two methods, its overall error is reduced to a certain extent, especially at some key time points, where the peak error is significantly reduced. This indicates that using the method provided by this invention to estimate time calibration parameters and calibrate sensor data can effectively improve the accuracy of multi-sensor data fusion, making the target state estimation closer to the true value.
[0066] Multi-target fusion method with known time calibration parameters: The green solid line represents the ideal case where the time calibration parameters are known. Its error is relatively low and relatively stable, demonstrating that under ideal time synchronization conditions, multi-sensor data fusion can achieve good target tracking results. It also shows that methods considering time calibration parameter estimation are approaching the ideal situation, verifying the effectiveness and necessity of time calibration parameter estimation.
[0067] Figure 2 This is a flowchart illustrating the method for constructing the posterior probability density expression of time calibration parameters using Gaussian mixture approximation, as provided by the present invention.
[0068] Step 201: Use the generalized covariance cross-fusion criterion to determine the initial posterior probability density expression of the time calibration parameters.
[0069] The initial posterior probability density expression for the time calibration parameters includes the prior probability density function of the time calibration parameters, and the likelihood function between the first multi-objective probability density function and the second multi-objective probability density function with respect to the time calibration parameters.
[0070] The posterior probability density expression for the time calibration parameter is given by the following formula: ; in, To integrate weights, To track the state space of the target, For time calibration parameters, Represents the set of all possible time calibration parameters. Let be the prior probability density function of the time calibration parameters. and Let represent the first multi-objective probability density function and the second multi-objective probability density function, respectively.
[0071] Let be the likelihood function, used to represent the joint likelihood between the first and second multi-objective probability density functions given time calibration parameters. The formula for calculating the likelihood function is shown below.
[0072] .
[0073] Step 202: Using the Gaussian mixture approximation technique and the probability hypothesis density filter, determine the Gaussian mixture exponential form of the likelihood function.
[0074] In some embodiments, a probability hypothesis density filter can be used to determine the mathematical expression of the likelihood function; and the Gaussian mixture approximation technique can be used to express the mathematical expression of the likelihood function in Gaussian mixture exponential form.
[0075] In the specific implementation process, the first multi-objective probability density function can be processed by the probability hypothesis density filter to obtain the first intensity function; the second multi-objective probability density function can be processed by the probability hypothesis density filter to obtain the second intensity function; the first intensity function and the second intensity function can be used to represent the likelihood function, resulting in the mathematical expression of the likelihood function as shown below.
[0076] ; in, For time calibration parameters, , , For the first intensity function, This is the second intensity function under the time calibration parameters. To track the state variables of the target; For time offset; Weighting coefficients; For the reference sensor to observe the target at the time point, The time point at which the sensor to be registered observes the same target.
[0077] Assuming time offset The value is relatively small, resulting in the following mathematical expression for the likelihood function: ; in, ; in, For the first intensity function, This is the second intensity function under the time calibration parameters. To track the state variables of the target; These are variables related to time calibration parameters; These are the weighting coefficients; Then, using the Gaussian mixture approximation technique, the mathematical expression of the likelihood function is represented in Gaussian mixture exponential form. The specific process is shown in the following formula: First, based on the Gaussian mixture (GM) approximation technique, we give... Approximate expression: ; in, and Let represent the number of Gaussian components associated with the first and second multi-objective probability density functions, respectively. δ is a weighting coefficient, a function of δ. This coefficient is used to adjust the contribution of each Gaussian component combination in the summation process, and its value may depend on the time calibration parameter δ, reflecting the importance of different Gaussian components under different time calibration conditions. It is a multivariate Gaussian distribution. The variable is a Gaussian distribution and is related to the time calibration parameter δ and the Gaussian component combination (a,b) of the first and second multi-objective probability density functions; The covariance matrix, also related to the time calibration parameter δ and the Gaussian component combination (a, b), describes the dispersion of variables and the correlation between variables in a Gaussian distribution. Wherein: ; ; ; in, and : These are the original mixing coefficients of the a-th and b-th Gaussian components in the first and second multi-objective probability density functions, respectively, reflecting the inherent weight of each Gaussian component in its respective probability density function; ω is the fusion weight coefficient, which usually ranges from 0 to 1 and is used to balance the contributions of the first and second multi-objective probability density functions in the fusion process; The prior probability associated with the time calibration parameter δ reflects the impact of time calibration on sensor data fusion; This is a kernel function used to adjust the calculation of weighting coefficients to ensure their rationality. The covariance matrix of the first multi-objective probability density function And related to the fusion weight ω, the weights of the first multi-objective probability density function are adjusted; To adjust the weights of the second multi-objective probability density function, where It is the covariance matrix after adjustment by the time calibration parameter δ; m1,a: the mean vector of the a-th Gaussian component in the reference sensor, representing an estimate of the target state by the reference sensor; The mean vector of the b-th Gaussian component of the second multi-target probability density function after adjustment by the time calibration parameter δ reflects the estimation of the target state by the sensor to be registered after considering time calibration. Let be the covariance matrix of the a-th Gaussian component of the first multi-objective probability density function. It is the covariance matrix of the b-th Gaussian component in the second multi-objective probability density function after being adjusted by the time calibration parameter δ.
[0078] Through mathematical simplification Further characterization is as follows: ; The Gaussian mixture exponential form of the likelihood function is expressed by the following formula: ; Where a and b are the index numbers, Let be the number of Gaussian components corresponding to the first multi-objective probability density function. Let be the number of Gaussian components corresponding to the second multi-objective probability density function. These are weighting coefficients used to balance the contributions of different terms to the objective function. It follows a normal distribution. For time calibration parameters, The mean of the Gaussian components is given. Let V be the variance of the Gaussian components.
[0079] Step 203: Based on the likelihood function and prior probability density function in Gaussian mixture exponential form, obtain the posterior probability density expression of the time calibration parameter in Gaussian mixture exponential form.
[0080] Assume the prior probability density function of the time calibration parameters Having the form of a Gaussian mixture exponent, the posterior probability density expression of the time calibration parameter in Gaussian mixture exponent form can be obtained from the likelihood function in Gaussian mixture exponent form and the prior probability density function of the time calibration parameter, as shown below.
[0081] ; in, This represents the normal probability density function with time calibration parameters as variables. This is the parameter vector for time calibration parameters. Let be the mean of the i-th distribution. Let be the variance of the i-th distribution. The weights for each Gaussian component.
[0082] The distributed multi-target fusion tracking device based on time calibration provided by the present invention will be described below. The distributed multi-target fusion tracking device based on time calibration described below can be referred to in correspondence with the distributed multi-target fusion tracking method based on time calibration described above.
[0083] Figure 8 This is a schematic diagram of the structure of the distributed multi-target fusion tracking device based on time calibration provided by the present invention.
[0084] like Figure 8 As shown, the time-calibrated distributed multi-target fusion tracking device 800 includes the following modules: The construction module 810 is used to construct the posterior probability density expression of the time calibration parameter based on the generalized covariance cross-fusion criterion, using Gaussian mixture approximation technology and probability hypothesis density filter; wherein, the posterior probability density expression is in the form of Gaussian mixture exponential function; the time calibration parameter is used to align the first multi-objective probability density function and the second multi-objective probability density function in the time dimension; the first multi-objective probability density function is obtained by filtering the data collected by the reference sensor, and the second multi-objective probability density function is obtained by filtering the data collected by the sensor to be registered.
[0085] Selection module 820 is used to select the mean of the Gaussian component with the largest weight from multiple Gaussian components of the posterior probability density expression as the estimate of the time calibration parameter.
[0086] The synchronization module 830 is used to align the second multi-objective probability density function with the first multi-objective probability density function in time based on the estimated value of the time calibration parameter, so as to obtain the calibrated second multi-objective probability density function.
[0087] The fusion module 840 is used to fuse the first multi-target probability density function and the calibrated second multi-target probability density function based on the generalized covariance cross-fusion criterion to obtain the multi-target state estimate of the tracked target.
[0088] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9As shown, the electronic device may include: a processor 910, a communications interface 920, a memory 930, and a communications bus 940, wherein the processor 910, the communications interface 920, and the memory 930 communicate with each other through the communications bus 940. The processor 910 can call logic instructions in the memory 930 to execute a time-calibrated distributed multi-target fusion tracking method. This method includes: constructing a posterior probability density expression for a time calibration parameter based on a generalized covariance cross-fusion criterion, using Gaussian mixture approximation techniques and a probability hypothesis density filter; wherein the posterior probability density expression is in the form of a Gaussian mixture exponential function; the time calibration parameter is used to align a first multi-target probability density function and a second multi-target probability density function in the time dimension; the first multi-target probability density function is obtained by filtering data collected by a reference sensor, and the second multi-target probability density function is obtained by filtering data collected by the sensor to be registered; selecting the mean of the Gaussian component with the largest weight from multiple Gaussian components of the posterior probability density expression as an estimate of the time calibration parameter; aligning the second multi-target probability density function with the first multi-target probability density function in time according to the estimated value of the time calibration parameter to obtain a calibrated second multi-target probability density function; and fusing the first multi-target probability density function and the calibrated second multi-target probability density function based on the generalized covariance cross-fusion criterion to obtain a multi-target state estimate of the tracked target.
[0089] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0090] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the distributed multi-target fusion tracking method based on time calibration provided by the above methods. This method includes: constructing a posterior probability density expression for time calibration parameters based on a generalized covariance cross-fusion criterion, using Gaussian mixture approximation techniques and probability hypothesis density filters; wherein the posterior probability density expression is in the form of a Gaussian mixture exponential function; and the time calibration parameters are used to achieve alignment between the first multi-target probability density function and the second multi-target probability density function in the time dimension. The first multi-target probability density function is obtained by filtering the data collected by the reference sensor, and the second multi-target probability density function is obtained by filtering the data collected by the sensor to be registered. The mean of the Gaussian component with the largest weight is selected from the multiple Gaussian components of the posterior probability density expression as the estimated value of the time calibration parameter. Based on the estimated value of the time calibration parameter, the second multi-target probability density function is aligned with the first multi-target probability density function in time to obtain the calibrated second multi-target probability density function. Based on the generalized covariance cross-fusion criterion, the first multi-target probability density function and the calibrated second multi-target probability density function are fused to obtain the multi-target state estimate of the tracked target.
[0091] Furthermore, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the time-calibrated distributed multi-target fusion tracking method provided by the methods described above. This method includes: constructing a posterior probability density expression for time calibration parameters based on a generalized covariance cross-fusion criterion, utilizing Gaussian mixture approximation techniques and a probability hypothesis density filter; wherein the posterior probability density expression is in the form of a Gaussian mixture exponential function; the time calibration parameters are used to align a first multi-target probability density function with a second multi-target probability density function in the time dimension; the first multi-target probability density function is determined by the parameters... The data collected by the sensor is filtered to obtain the second multi-target probability density function, which is obtained by filtering the data collected by the sensor to be registered. The mean of the Gaussian component with the largest weight is selected from the multiple Gaussian components of the posterior probability density expression as the estimated value of the time calibration parameter. Based on the estimated value of the time calibration parameter, the second multi-target probability density function is aligned with the first multi-target probability density function in time to obtain the calibrated second multi-target probability density function. Based on the generalized covariance cross-fusion criterion, the first multi-target probability density function and the calibrated second multi-target probability density function are fused to obtain the multi-target state estimate of the tracked target.
[0092] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0093] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A distributed multi-target fusion tracking method based on time calibration, characterized in that, include: Based on the generalized covariance cross-fusion criterion, a posterior probability density expression for the time calibration parameter is constructed using Gaussian mixture approximation and probability hypothesis density filter. The posterior probability density expression is in the form of a Gaussian mixture exponential function. The time calibration parameter is used to align the first multi-objective probability density function and the second multi-objective probability density function in the time dimension. The first multi-objective probability density function is obtained by filtering data collected from a reference sensor, and the second multi-objective probability density function is obtained by filtering data collected from the sensor to be registered. The mean of the Gaussian component with the largest weight is selected from the multiple Gaussian components of the posterior probability density expression as the estimated value of the time calibration parameter. Based on the estimated value of the time calibration parameter, the second multi-objective probability density function is aligned with the first multi-objective probability density function in time to obtain the calibrated second multi-objective probability density function; Based on the generalized covariance cross-fusion criterion, the first multi-target probability density function and the calibrated second multi-target probability density function are fused to obtain the multi-target state estimate of the tracked target.
2. The distributed multi-target fusion tracking method based on time calibration according to claim 1, characterized in that, The method for constructing the posterior probability density expression of the time calibration parameters based on the generalized covariance cross-fusion criterion, using Gaussian mixture approximation techniques and probability hypothesis density filters, includes: The initial posterior probability density expression of the time calibration parameter is determined using the generalized covariance cross-fusion criterion; wherein the initial posterior probability density expression of the time calibration parameter includes the prior probability density function of the time calibration parameter, and the likelihood function between the first multi-objective probability density function and the second multi-objective probability density function with respect to the time calibration parameter. The Gaussian mixture approximation technique and the probability hypothesis density filter are used to determine the Gaussian mixture exponential form of the likelihood function; Based on the likelihood function in Gaussian mixture exponential form and the prior probability density function, the posterior probability density expression of the time calibration parameter in Gaussian mixture exponential form is obtained.
3. The distributed multi-target fusion tracking method based on time calibration according to claim 2, characterized in that, The process of determining the Gaussian mixture exponential form of the likelihood function using Gaussian mixture approximation techniques and probability hypothesis density filters includes: The mathematical expression of the likelihood function is determined using a probability hypothesis density filter; Using the Gaussian mixture approximation technique, the mathematical expression of the likelihood function is represented in Gaussian mixture exponential form.
4. The distributed multi-target fusion tracking method based on time calibration according to claim 3, characterized in that, The step of using a probability hypothesis density filter to determine the mathematical expression of the likelihood function includes: After processing the first multi-objective probability density function using a probability hypothesis density filter, a first intensity function is obtained; After processing the second multi-objective probability density function using a probability hypothesis density filter, the second intensity function is obtained; Using the first intensity function and the second intensity function, the likelihood function is expressed, and its mathematical expression is obtained.
5. The distributed multi-target fusion tracking method based on time calibration according to claim 4, characterized in that, The likelihood function is expressed by the following formula: ； in, ； in, For the first intensity function, This is the second intensity function under the time calibration parameters. To track the state variables of the target; These are variables related to time calibration parameters; These are the weighting coefficients; The Gaussian mixture exponential form of the likelihood function is expressed by the following formula: ； Where a and b are the index numbers, The number of Gaussian components corresponding to the first multi-objective probability density function. The number of Gaussian components corresponding to the second multi-objective probability density function. These are weighting coefficients used to balance the contributions of different terms to the objective function. It follows a normal distribution. For time calibration parameters, The mean of the Gaussian components is given. Let V be the variance of the Gaussian components.
6. The distributed multi-target fusion tracking method based on time calibration according to claim 1, characterized in that, The method of fusing the first multi-target probability density function and the calibrated second probability hypothesis density filter multi-target probability density function based on the generalized covariance cross-fusion criterion to obtain the multi-target state estimate of the tracked target includes: Based on the generalized covariance cross-fusion criterion, the first multi-target probability density function and the calibrated second multi-target probability density function are fused to obtain the Gaussian component of the multi-target state association of the tracked target. From the Gaussian components associated with the multi-target state, the mean of the Gaussian components with a weight value greater than a preset threshold is selected as the multi-target state estimate of the tracked target.
7. A distributed multi-target fusion tracking device based on time calibration, characterized in that, include: A construction module is used to construct a posterior probability density expression for time calibration parameters based on the generalized covariance cross-fusion criterion, using Gaussian mixture approximation techniques and probability hypothesis density filters; wherein, the posterior probability density expression is in the form of a Gaussian mixture exponential function; the time calibration parameters are used to align the first multi-objective probability density function and the second multi-objective probability density function in the time dimension; the first multi-objective probability density function is obtained by filtering the data collected by the reference sensor, and the second multi-objective probability density function is obtained by filtering the data collected by the sensor to be registered; The selection module is used to select the mean of the Gaussian component with the largest weight from multiple Gaussian components of the posterior probability density expression as the estimate of the time calibration parameter. The synchronization module is used to align the second multi-objective probability density function with the first multi-objective probability density function in time based on the estimated value of the time calibration parameter, so as to obtain the calibrated second multi-objective probability density function. The fusion module is used to fuse the first multi-target probability density function and the calibrated second multi-target probability density function based on the generalized covariance cross-fusion criterion to obtain the multi-target state estimate of the tracked target.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the time-calibrated distributed multi-target fusion tracking method as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the time-calibrated distributed multi-target fusion tracking method as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the time-calibrated distributed multi-target fusion tracking method as described in any one of claims 1 to 6.