Object Tracking Algorithm Research Based On Particle Filter

This paper, through analyzing and comparing the advantages and disadvantages of common video tracking algorithms and the particle filter algorithm, mainly studied the application of the latter algorithm in video tracking and its technical problems, and illustrate its advantages. Particle resampling and multi-feature fusion were also proposed to solve the existing problems and improve the algorithm.The paper gave an introduction to the concept and characteristics of video tracking algorithms, and explained the classification of video detecting algorithms. It mainly introduced four algorithms, namely background differencing, coterminous frames differencing, optical flow and particle filter algorithm, and elaborated on the last algorithm, detailing its theoretical basis and implementation. It introduce the relationships among Bayesian filter, kalman filter and particle filter, and the basic theory about particle filter algorithm.It first presented the application of the particle filter algorithm, then described steps and ways to implement it, including tactics it used to select and integrate features, and at last touched upon existing problems with this algorithm. The fourth chapter would be largely devoted to resampling and multi-feature fusion.Based on researches, this paper improved the algorithm, proposed resampling and multi-feature fusion, and conducted simulation experiments and comparison to verify the feasibility and performance of the algorithm. It intensively examined thesampling algorithm and multi-feature fusion in the algorithm. Since there are still some deficiencies with the particle filter algorithm, the study proposed the new resampling algorithm, which can meet the requirements in terms of stability and speed of particle tracking. The multi-feature fusion method can automatically adjust multi-feature weights according to accuracy, stability and robustness. In the end, the study compared the proposed multi-feature fusion-based particle filter algorithm with other algorithms, analyzed stimulation data and proved the feasibility of the proposed algorithm.