Object recognition and detection play important roles in various computer vision applications. When images contain different types of objects, detection of an object-of-interest (OOI) from multiple stacking objects becomes a difficult task to be handled using a monocular camera. In this paper, a novel automatic target selection and tracking algorithm is proposed to address this issue efficiently. The proposed method first uses keypoint correspondences to compute control points of targets appeared in incoming images. Next, each OOI in captured images is separated from multiple stacking objects randomly placed in a box using mean shift clustering approach. Finally, a template-based visual tracking method is used to locate and track center position of the top OOI in the box. When implemented on an Intel Core i5-4440 3.1GHz platform, the proposed algorithm achieves real-time performance about 30 frames per second at 640x480 image resolution in the experiments.
ICIC express letters. Part B, Applications : an international journal of research and surveys 7(5), p.1135-1140