Visual Tracking using Corner based Centrist Descriptor with a Robust Localization Algorithm

In this paper an algorithm for object tracking in the visual domain based on a novel localization method is proposed. First a part of the search area, preferably the interest points is chosen. The proposed approach drastically speeds up the process of tracking, meanwhile the intensity histogram and Centrist descriptor which is known for good coding capability of small patches of an image will be used for target’s description. In order to increase the accuracy of the descriptor, this descriptor is applied to small blocks of image to encode most of the image around the target’s interest points. By providing the description of object’s interest points, a 1-NN classifier is used to distinguish the corresponding target’s interest points in each frame. Given the matched corresponding interest points, a convolution problem is formulated to detect the center of the target. Experiments on a challenging dataset against several state-of-theart methods demonstrate the efficiency of the proposed algorithm.