1Science and Research Branch, Islamic Azad University
3Sahand University of Technology
Hand gestures are powerful way for human communications. The proposed method is effectively combined of the following steps to detect hand gesture. First, . Then two methods of extracting Haar-like features and histogram of oriented gradient features (HOG) are applied and combination of those features forms a special feature vector. By adding advance half size Haar-like features to the basic Haar-like features and homomorphic filtering performance of Haar-like features improved. Also by applying the new Tan and Trigger preprocessing before HOG, sensitivity of lightening conditions have been reduced. Finally, linear multi-class support vector machine classification is used. The system is tested on Massey university American sign language (ASL) numeric and alphabetic hand gesture datasets, and system have been successfully able to recognize hand gestures with the recognition rate of 98% on numeric ASL and the recognition rate of 93% on alphabetic ASL. In addition recognition rate was 96% on National University of Singapore (NUS) hand gesture dataset with crowed backgrounds and variant lightening conditions.
Hand Gesture Recognition; American Sign Language (ASL); Haar-like Features; Histogram of Oriented Gradients (HOG); Support Vector Machine (SVM)