The Key Technology Researches On Robot Vision-based Hand Gesture Recognition

By analyzing the features of gesture and theory of neural network, and based on pyramid LK theory, FERN classifer theory, P-N learning theory, as well as theories of motion trajectory continuity, the dissertation studies gesture recognition and relevant key technologies, that is, target gesture tracking technology, P-N learning samples growth and pruning technology and new neural network algorithm based on gesture trajectory model.The main research work is as follows:1.In order to handle with the difficulty of recovering after a failure target tracking and that the target is susceptible to influence by face,color of skin and arms which will appear during the process of tracking the local area of hand, the paper presents an improved multi-objective local area tracking algorithm which is based on original TLD tracking algorithms.Based on FERN classifier, the paper use the method of color characteristics intersection coedfficient to improve the first stage of variance classifier for detection modul.And the paper also presents an adaptive method of coordinate system to reduce the slide area of scan window in detection module.2.Based on the above tracking results,aiming at to the fuzziness and accuracy of gesture recognition, the paper emploies a clockwork recurrent neural network algorithm to recognize dynamic gesture in human-computer interaction.Then the paper also verify and anslysis deeply the factors which affect the CW-RNN model in experiments.And by comparing with the multilayer RNNs and three layer RNNs,the paper demenstrates the superiority of proposed model of trajectory template of gesture recognition.