Recognizing Hand Gestures Using Local Features: A Comparison Study

Interest point approaches that extract local features from images are commonly used in human action recognition field. In this paper, a comparison study is performed in which different interest point approaches are used. Each approach is discussed with its advantages and drawbacks. Common keypoint extractors like scale invariant features transform (SIFT), speeded up robust features (SURF), etc. are used in context to human hand gestures recognition. In human-robot interaction, efficiency is important in any recognition task along with recognition rate. Hence in this work, performance of 8 different versions of keypoints are evaluated in terms of recognition rates along with their robustness and efficiency with respect to time. SIFT features show best recognition results while SURF and maximally stable extremal regions features (MSER) show better efficiency.