Optimize particle filter tracking algorithm by features fusion and variable noise covariance

A particle filter target tracking algorithm is proposed to solve tracking problems in complex scene. Used united histogram according to weighted background as in Eq. 7 to describe gradient direction and grayscale features information of target, adjusted features weights adaptively in the observational model based on features´ dependability as in Eq.9, and combined the observational model with particle filter, designed a variable state noise covariance to moving model as in Eq.14 to control the main seeking area of particles to adopt tracking conditions. The experimental result given by Fig. l~Fig. 3 shows using features fusion algorithm precedes methods based on single feature.