In order to solve the problem resulted from tracking drift of single reference color histograms，a sort of particle filtering algorithm based on DSPCA(decomposition-sparse-principal component-analysis) is presented. The image set with complementary is accessed by DSPCA，which is fused by means of the new similarity as the new object model. Compared with the basic particle filtering，visual tracking decomposition and adaptive multiple cues fusion tracking algorithm in challenging scenarios，the presented method can robustly track the object in fast motion and the object which is mostly blocked.