Visual target tracking based on multi-view feature fusion with online multiple instance learning

摘要

The tracking methods under the online multiple instance learning (MIL) framework always use single channel's information or transform the RGB image to gray image for color video tracking, which may cause the information loss of the color image. In addition, the MIL tracker only use the haar-like feature to construct the target appearance and ignore the different samples have different importance to the tracking results, which is argued to susceptible to interference with complex background. Therefore, this paper presents an online visual target tracking method based on multi-view feature fusion with joint online multiple instances learning to overcome the disadvantages of the MIL tracker. Firstly, we extract the color histogram feature and haar-like features to construct the two type classifiers. Secondly, we compute the bag probability by summing the instance probability in the process of updating the classifiers and construct the strong classifiers. Finally, the object location is achieved by fusing the two type strong classifiers. The experimental results show that the proposed method performs better than several other tracking methods like IVT, MIL, OAB and WMIL trackers with smaller central position error on challenging color video sequences.