A hybrid algorithm combined color feature and keypoints for object detection

Object detecting methods based on local keypoints have shown stable and effective performance to detect object in clutter and occlusion environment, but these methods usually ignore color region information which is useful for object to distinguish it from background and other objects. To improve the veracity and celerity of object detection under clutter, we present a new approach for object detection with the basic idea of multi-features unity. Firstly, we define a novel color region descriptor called object dominant color set (ODCS). Then use an indirect approach combining ODCS region-based detection and scale invariant feature transform (SIFT) keypoint-based detection in an ordinal coarse-to-fine manner. At last, introduce a voting mechanism to decide the final detection result. The proposed approach has both the celerity of color region-based detector and veracity of keypoint-based detector concurrently, and is more effective to detect objects of few keypoints. The experimental data illuminate the potential of the proposed approach.