| Today, science and technology is very mature, human-computer interaction has been gradually known by people and has been an indispensable part of people’s daily life. Hand gesture recognition is nature and simple for communication. In recent years, Hand gesture recognition has become a research focus of human-computer interaction.In this paper, multi-objective tracking, dynamic gesture recognition in monocular vision and the classification and dimension reduction algorithm based on manifold learning are explored, then the paper firstly put forward a multi-objective tracking algorithm which is based on skin color cue and oriented k-Dops. The algorithm can not only track multi hands, but also precisely identified and track hand even when occlusions exist. Secondly, the paper put forward an effective gesture recognition and reconstruction algorithm. The algorithm uses Locality Preserving Projections (LPP) to learn the manifold, and then recognize and reconstruct the gesture from multi-views. Firstly, we use some standard motion to learn the manifold by LPP in order to get the embedding spaces. In order to classify and estimate more precisely by embedding spaces, we brought forward a manifold space positioning algorithm. The algorithm not only can find out which manifold the gesture belongs but also can recognize and reconstruct the gesture perfectly.Experiments prove that the algorithm can track multi hands and recognize and reconstruct the gesture in time in monocular vision precisely and robustly. |
