Object tracking using color-feature guided network generalization and tailored feature fusion

In this paper a particle filter based algorithm for color-guided object tracking is proposed to solve problems such as object drifting and lost in complex environment. Firstly, strong object and weak object are differentiated based on color feature relevance between object and background. Secondly, a self-adaptive object model is constructed by object status with tailored features that include CNN feature produced by defined network structure with fixed kernel functions describing object's general property, HOG feature describing object's specific property and color feature. Then the searching strategy of spatial consistency under the guidance of color feature is applied to approach tracking result. In the end, the bounding box of object is optimized by use of the mapping size of mathematical space. The proposed algorithm reduces background noise and improves tracking accuracy of objects with changing appearance. And the effectiveness of the proposed algorithm is validated by final result of the experiment.