Real-time Moving Target Tracking by Integrating Multiple Model into Particle Filtering
Linlin Wu · Journal of Highway and Transportation Research and Development · 2010
A novel algorithm for integrating multiple model into particle filter,MMGPF,was proposed for moving target tracking and applied to pedestrian and vehicle tracking.It has some innovative characters:(a) The proposal probability distribution was optimized by incorporating the outputs of Camshift and AdaBoost into the IDPF framework,and the framework of particle filtering was improved,leading to efficiency improvement of the particle filter sampling and dramatically reduction of particle numbers without affecting the tracking performance.(b) Using two descriptors,the HOG descriptor and HSV color histogram,to enhance observation model.(c) Using two kinds of method to speed up the proposed algorithm.Owing to integrating multiple models,the MMGPF implicitly handles the difficulties of tracking caused by object occlusion,object disappearance and reappearance,illumination variation and background clutters.It is demonstrated through several real tracking tasks that the new method performs well in both tracking robustness and computational efficiency.