Object tracking based on multi-feature fusion and motion prediction

In order to strengthen the object identify ability of the tracking algorithm in image sequences and deal with occlusion, an object tracking method, mixed with color histogram, scale invariant feature transform (SIFT) and the motion prediction based on α-β-γ filter, is put forward in this paper. Firstly, the mean shift and SIFT algorithm are used on the candidate object in the current frame of image, to obtain the object model and the SIFT feature, which are then compared with those of the previous frame for computing the similarity coefficients of the mean shift algorithm and the SIFT feature. Then the object's real position is calculated by comparing and iterating these two coefficients and the position is used as the basis for updating the α-β-γ filter. Finally, the object position in the next frame is predicted by the α-β-γ filter. With occlusion, the predicted value from α-β-γ filter is used to keep the tracking continuity. Experimental results show that the algorithm has good real-time performance and small errors while tracking object in the sequence. © 2011 by Binary Information Press.