The aim of this paper is to explore and recognize hand gestures against complex backgrounds and lighting conditions. Mostly, previous approaches use off-line recognition for the hand motion, which make any gesture system not realistic and not suitable for on-line applications. Furthermore, 3D dynamic features are further extracted from the hand gesture path based on stereo strategy. Thus, the extracted features of dynamic affine-invariants such as location, orientation and velocity are derived

more »

... rom 3D spatio-temporal hand motion. After that the feature vectors are constructed by quantization and then employed to multi-class Support Vector Machine for training or texting. An application of isolated gestures from alphabet characters (A-Z) and numbers (0-9) are trained and tested with superior performance. Furthermore, the results are promising on several video samples in different situations.