Particle Filter Algorithm And Dsp-based Target Tracking System

With the development of technology, target tracking has become a hot topic in the field of computer and communication technology, it has a very wide range of applications. In the conditions of fast-moving, background blocking and noise, precise tracking is the focus of scientific researcher. For example, the tracking of missiles, aircraft and other target in the area of military, In the area of traffic, it can process real-time dynamic monitoring at the crowded places, such as highway, crossway, and subway.The emergence of DSP microprocessors makes digital signal processing technology has been rapid development. DSP apply in very broad fields as it has high integration, high-speed CPU and large capacity memory. We can imagine that with the increasing requirements of various sectors and the further development of various technologies, tracking of moving targets with DSP will be more widely used by everyone. However, due to a short history and come down to numbers of new subjects such as computer vision, image processing, pattern recognition, artificial intelligence and automatic control, the theory is far from perfect, some important issues unresolved, new methods and techniques have to be developed, so target tracking is a practical value of both theoretical issues.DSP processor can achieve many algorithms, the author make depth study of the Particle filter algorithm,Particle filter is a Bayesian filter which is recursive and base on transmitting sample set, it can keep multiple hypotheses and use random target motion model to predict the location, so it can better deal with the background noise and make self-recovery when loss or failure in target tracking case.However, the tracking accuracy of particle filter is related to the number of particles, excessive number of particles will lead to excessive calculation.In the course of the study, the author came up with the idea of combining with other algorithms.the Mean Shift algorithm is based on the increase in non-parametric density gradient algorithm, through several iterations to find the target location to achieve the target tracking, it need a small calculation and be easy to achieve but it is easily lost to tracking of small targets and can not be self-recovery when the target is blocking out. The mean shift algorithm which combines the particle filter, gathered samples of the particles, get more samples with high weight, remove the samples which with tiny contribution to the tracking, reduce the number of samples, thus greatly enhance the real-time algorithm.