Object Tracking Based on Channel Attention

Object tracking is one of the important research topics in computer vision. Despite the great progress in this area, effectively and efficiently tracking object in videos still remains challenging especially in scenarios of rapid movement of objects, illumination changes, appearance of similar objects in the background. In this work, we propose an object tracking method by improving the state-of-the-art tracker MDNet based on channel attention, which can distinguish among similar objects by suppressing the background and highlighting the object. We integrate the channel attention module and group normalization into MDNet network. To validate the effectiveness of the proposed tracker, we compare it with a number of existing state-of-the-art trackers in terms of success rate and precision on OTB-100, OTB-50 and CVPR 2013 datasets. The test results have demonstrated the effectiveness and improvement of the proposed tracking algorithm.