Probability Hypothesis Density Based Multi-sensor Multi-target Tracking Algorithms

With the increasingly complex battlefield environment,the tracking precision and reliability of single sensor multi-target tracking system are unable to meet the requirements of modern warfare.Meanwhile,with the development of information processing technology and sensor technology,multi-sensor tracking system has been used to estimate the motion state of targets.The aim of multi-sensor tracking system is to reduce the influence of environment on multi-target tracking and perform better than a single sensor.It is a difficult and hot issue that how to fuse measurement information of multiple sensors effectively in the field of multi-sensor multi-target tracking.Recently,great attention has been attached to the multi-sensor multi-target tracking method based on the random finite set(RFS)theory.As a new approach to solve multi object tracking in complex environment,this method adopts serial system structure and models the sets of multi-target states and measurements as random finite sets.In this way,the multi-sensor multi-target state estimation can be handled by multiple single sensor Bayes filters.Also,the data associations in multi-sensor fusion and traditional target tracking algorithm can be avoided.The multi-sensor multi-target tracking algorithms based on RFS theory and probability hypothesis density(PHD)filter are studied deeply in this dissertation.The main contributions of the dissertation are given as follows:1.The problem of sensor update order is caused by probability of detection.In an iterated corrector PHD(IC-PHD)filter,the tracking performance is largely depended on the probability of detection of the last update sensor.Targets are easily undetected by the tracking system when the probability of detection of the last update sensor is low.To deal with this problem,an improved algorithm based on the Gaussian implementation of IC-PHD filter is proposed.The architecture of the improved algorithm is similar to that of the original algorithm.However,the probabilities of detection and miss-detection of each Gaussian component are obtained by fusing the probabilities of detection and miss-detection of multiple sensors.The simulation results indicate that the improved algorithm can reduce the impact of probability of detection and weaken the influence of sensor order.2.The false targets are caused by miss-detection.In a product multi-sensor PHD(PM-PHD)filter,it needs to calculate the infinite sum in the modified coefficient.The infinite sum is computationally problematic,because the value of each term in it is positive.Therefore,an approximate approach for this problem is proposed.Based on the convergence analysis of the infinite terms,the infinite sums are replaced by the sums of a few representative finite terms.Additionally,in the Gaussian implementation of the PM-PHD filter,if a target is undetected,false targets may exist.To address this problem,a redistribution method for the weights of Gaussian components is proposed.The simulation results show that the redistribution method can avoid false targets and miss-detection,and improve the performance of the filtering algorithm effectively.3.The incorrect target weight estimation is caused by not making good use of measurement information.In the IC-PHD filter,the weight of measurement subsets generated by the measurement partition may be too large or too small.Because the weight calculation method is difficult to make full use of the measurement information of multiple sensors.In this regard,an improved weight calculation method is presented.In this method,the weight of measurement subset is divided into two parts.One is mainly used to improve the low weight caused by miss-detection,the other is mainly used to handle the excessive weight caused by false alarms.The simulation results show that the proposed method can effectively improve tracking accuracy and robustness of filtering algorithm.4.The incorrect target state estimation is caused by multi-sensor system architecture.The tracking performance of the PM-PHD filter is easily affected by the sensor order and probability of detection.Because the posteriori PHD is still updated by the iterated corrector process.This may lead to an incorrect target state estimation.To deal with problem,an improved algorithm is proposed to estimate the target number and target state separately is proposed.In the improved algorithm,the PHD updating process consists of two parts.One is that the number of targets is estimated directly by the expectation of cardinality distribution,the other is that the state of targets is estimated by the normalized PHD.The simulation results indicate that the improved algorithm can avoid the drawbacks of the original algorithm and weaken the impact of system performance on the tracking results.The above four sections are complement each other and comprise a systematic series of multi-sensor multi-target tracking algorithms.They provide new views and technical support for the multi-target tracking in complex environment.