Hand Gesture Recognition Based on Cascading of Multiple Features

Hand recognition using gestures is gaining high attention in the era of human computer interaction. Gestures, an aphonic body language play prominent role to convey various messages in daily communication. Hand gesture recognition is proposed based on serial cascading of multiple features, motion, location, and shape components extracted from both segmented semantic data and entire gesture sequence. Enhanced principal component analysis (PCA) extracts the motion component by analyzing space interdependency among the neighboring motion energy histogram bins. The gesture location component is extracted through particle-based weighted dynamic time wrapping (PWDTW) while spatio-temporal interest points (STIP) of possible gestures is employed for shape component extraction. The proposed system performance is evaluated with several matchers, namely, Euclidean distance, Hamming distance, Extended jaccard coefficient (EJC), least cost methods of minimum cost matcher (MCM) and optimal cost region matcher (OCRM). A low computational recognition time is observed from the experiment results when multiple gesture features are fused sequentially in contrast with single feature of hand gesture.