An Objects Detecting and Tracking method based on MSPF and SVM

Considering that the robust real-time tracking of non-rigid objects is difficult to realize, We present an objects detecting and tracking method based on mean-shift particle filter (MSPF) and support vector machine (SVM). The proposed algorithm uses the mean-shift vector of the tracking object to update the state transition matrix of particle filter algorithm, and we define the criterion of the particle degradation，to improve the conditions of degradation，the particles will be re-distributed as Gaussian distribution. Because of the real-time update of the particle motion parameters, the prediction accuracy of target motion parameters is improved. Under the condition of target conflicting and partially covering, the proposed algorithm is still tracking effectively. Apply SVM to relevance feedback of object detecting and tracking, the experiments results show that the method can overcome the shortness of the traditional methods, effectively improve the tracking speed and precision. The results of the experiment indicate that the average processing time per frame of the proposed algorithm is reduces by about 21% comparing with the classical ones，while the efficiency of particle increases by about 32%.