Research On Multi-sensor Multi-target Tracking Algorithm Based On Random Finite Set

With the continuous increase in the number of sensors,the complexity of sensor monitoring scenarios has further increased,and single-sensor systems can no longer meet the requirements of monitoring systems.In order to solve the problems of target tracking in complex scenarios,it is imperative to extend a single-sensor system to a multi-sensor system.In a multi-sensor system,different sensors cooperate to track the target in the scene.The performance of each sensor is different and the measurement information of the target detected is different,and the tracking accuracy is improved by comprehensive processing of the measurement information.Information fusion is the core content of data processing research in multi-sensor systems.By acquiring independent measurement data generated by sensors distributed in different positions in the scene,these data are fused to obtain effective information of the target,and then effective information is used for filtering.Complete tracking estimation of the status and number of targets.However,as the number of sensors in the sensor network increases,the measurement information detected by the sensors increases,resulting in a data explosion.In order to improve the ability of multi-sensor system multi-target tracking,domestic and foreign scholars have carried out extensive research on it in recent years,and due to the excellent characteristics of particle filter,the multi-sensor multi-target detection and tracking technology based on particle filter has been widely used in vehicle tracking,machine learning and modern medicine and other fields.Based on the stochastic finite set theory,this paper studies the fusion method of greedy algorithm and measurement,and obtains the following two research results:1.In order to solve the tracking problem of high-speed multi-maneuvering targets in complex scenes,the idea of interaction was combined with the idea of multi-sensors,and then the multi-sensor interactive Greedy potential balance probability hypothesis density(MS-IMM-Greedy-CPHD)filter was derived.In the prediction stage,the interactive multi-mode(IMM)algorithm is used to predict the state,potential distribution and motion model of the target in the CPHD filter simultaneously.In the update stage,the Greedy measurement partitioning mechanism was used to select the measurement subset and the quasi-partition,and the target state and potential distribution predicted by CPHD under different models were updated interactively through the quasi-partition subset.2.Aiming at the measurement division problem of multi-sensor and multi-target tracking,the greedy division mechanism of measurement fusion based on multi-bernoulli filter(MeMBer)is proposed.The partition mechanism firstly selects and fuses the measurements of different sensors,then selects the subset of the measurements and performs the operation of quasi-partition.In the update phase of the Multi-Bernoulli filter,the measurement information is operated by using the measurement fusion method combined with the greedy partitioning mechanism,and the posterior probability density of the target is updated by using the quasi-partitioned post-measurement subset.The simulation results show that the proposed fusion greedy partition mechanism can track the number and state of multiple targets stably and effectively.Compared with the measurement division of the greedy algorithm,the OSPA error of the tracking result is smaller and the potential estimation is more accurate when the same tracking performance is achieved.