Abstract: A particle filter object tracking algorithm based on dynamic feature fusion is proposed. The presented algorithm uses the complementary features, which are gray histogram and gradient histogram, to represent the object model. In the tracking procession, the confidence for each feature is adjusted according to the discrimination between the object and the background, and the object model is established and updated onlinely. The presented method can improve the accuracy of the object modeling and furthermore improve the accuracy of the particle filter tracking algorithm. Experimental results show that, in the representative object tracking scenes, the proposed algorithm can gain more accurate and more reliable tracking performance.