A new particle filter object tracking algorithm based on dynamic transition model
Jia Wei, Liu Hongjuan, Sun Wei, Pan Rong · 2016
At present, the main problems encountered in particle filter algorithm are low utilization rate and particle degeneracy. The new algorithm is based on the first order autoregressive model of particle movement. In the prediction step, the object mean-shift vector is used to update the state transition matrix (A) and the quantitative criterion (Neff) of particle degeneracy is also introduced. The particles are redistributed as per Gaussian distribution under the premise that degeneracy conditions are met. Due to the addition of real-time updating of particle movement parameters and the improvement in prediction accuracy of object movement parameters by the new algorithm, effective tracking can be achieved even though the object is in conflict and partially occluded. Experimental results show that when compared with the classical particle filter algorithm, the new algorithm has about 21% of reduction in average processing time per frame and about 32% of increase in particle utilization rate.