A fuzzy approach using directional features to hand gesture recognition for human–computer interaction

Despite their incredible desired properties of invariance, reliability and robustness, directional optical flow-based features still do not receive the attention they deserve in gesture recognition literature. Moreover, robust human hand pose estimation remains one of the most challenging tasks, because of its inherent difficulty caused by depth ambiguity and occluded hand joints. In this paper, an innovative fuzzy framework is proposed for hand gesture recognition and spotting, where an optimized local descriptor is constructed to model each moving gesture skeleton as a temporal series of fuzzy directional optical flow features. The Gentle AdaBoost (GAB) classifier is then trained on the normalized features for gesture classification, due to its ease of use and low parameter tweaking and less sensitivity to errors in the training process compared to other AdaBoost variants. When evaluated on a gesture dataset incorporating a large and diverse collection of gesture streams, the proposed model yields highly promising results (i.e., an accuracy of 98.39%) that compare very favorably with those reported in the literature, without sacrificing computational efficiency or stability. In addition, the experimental results validate the recognition model efficacy in spotting (extracting) meaningful hand gestures successfully from hand movements, while concurrently separating invalid gesture instances stemming from unintentional hand motions.