Video Object Tracking Based On Particle Filter

From the last century, with the accelerated pace of information, intelligence,a number of theoretical methods in image processing, computer vision and manyother areas are widely used. A variety of theoretical approaches have emerged intracking in recent ten years, such as mean-shift tracking algorithm, Kalmanfilter target tracking algorithms and particle filter tracking algorithm.Mean-shift tracking algorithm has aroused the concern of the many scholars andexperts because of its simple structure, easy implementation and so on, and is widelyused. However, Mean-shift tracking algorithm has some shortcomings, such asstrong dependence on color, fixed nuclear functionmutation window, and so on.Kalman filter is one type of Bayesian tracking theory, which can provide theoptimal minimum error estimate to linear, Gaussian system. Kalman filter trackingalgorithm is widly used and can successfully solve target tracking forthe linear Gaussian system. But Kalman filter tracking algorithm has significantlimitations to solve higher-dimensional, nonlinear, non-Gaussian target tracking.Particle filter was introduced in the1990s based on Monte Carlo method and therecursive Bayesian estimation as a new method of filtering; it has a unique advantagein dealing with non-linear, non-Gaussian parameter estimation. But the complexstructure of algorithms, large computational and slow computing speed limits itsapplication in real-time systems.TI TMS320DM6446is one of high-performance digital media processing chipfrom Texas Instruments Company, which contains a TMS320C64X+core and aARM926EJ core. C64X+DSP core’s operating frequency reach up to594M, whichcan effectively handle a large number of digital multimedia signals. The ARM corewhich runs embedded Linux system is able to effectively complete the process controloperations, and supply a stable platform for software developing.This article is written from the perspective of practical application; wefirst learned the theory of particle filter and got a good understanding, thenachieved the basic particle filter algorithm based on the particle filter theory. In orderto improve the accuracy of target tracking, this paper inproved the particle filteringalgorithm from several aspects, experimental results show that the improved particlefilter has a higher tracking accuracy. Then considering the shortcomings of complexstructure and high complexity, etc, this paper improved particle filter for therealization of the algorithm to optimize the structure and improve the efficiency of the algorithm, experimental results show that the optimized particle filter to carryout video tracking real-time.Finally, the improved particle filter tracking algorithmwas accomplished on TI’s DaVinci platform (DM6446), followed by someoptimization based on the platform.