Research On Target Tracking Algorithm Based On Particle Filter

The target tracking is widely used in many fields. The novel ideas or methods of theimproved target tracking accuracy have been more important. The target tracking is anestimation of the state of the dynamic system. In linear Gaussian system, the Kalman filteris the optimal estimation. However, in non-linear/non-Gaussian environment, the accuracyof the Kalman filters would be decreased quickly. The particle filter can be used effectivelyto estimate the nonlinear/non-Gaussian stochastic systems. The basic idea of particle filteris the probability density functions using the approximation of a series of discrete sampleswith particle weight. The sample is referred to as "particle". The principles of the particlefilter algorithm are deeply researched in this paper, and the main problems of the existingparticle filter algorithms are pointed out. Aimed at the key technology of the particle filter,the improvements on the basic particle filter are proposed in the following two respects：1. An improved resample algorithm called completely resample algorithm isproposed. If the effective number of particles is greater than the threshold, the particleswith large weights are copied simply, or the particles with large weights are not onlycopied, but also moved around. Compared with the several existing resample algorithms,the completely resample algorithm could ensure the variety of particles and enhance thestability of tracking. The simulation experiments show that the efficiency of the improvedalgorithm.2．Aimed at improving the density function, the double importance particle filteralgorithm is proposed. Two important density functions are selected in the algorithm.Firstly, the state transition function is selected as the important density function. The statetransition estimation is well utilized. Then, the likelihood function is selected as theimportant density function. It is a good use of the latest observations. Finally, the extractedparticles are summed up by weighting. The above method combined with the re-samplingalgorithm is called double importance particle filter algorithm. Compared with the standardparticle filter, the auxiliary particle filter, the likelihood particle filter, the new algorithmhas better tracking performance in greater noise environment. So the new algorithm couldenhance the stability and the reliability of the tracking.