The Locally Linear Embedding Algorithm Based On Improved Static Gesture Recognition And Dynamic Track

Recently, the motion analysis system for automatical human gesture recognition based on visual information have a unprecedented wide range of applications in interaction between computer and human beings, which includes mute hand gesture teaching with computer aid, demonstration of hand gesture directing TV set, many technologies in this research have already become more and more mature, special effects in movie production, animation making, medical research, electronic games and entertainment etc. And some design just be in concept development stage, Such as the AUDI AG this year display of the gesture recognition system equipped with the latest concept car. From the development of the research in all these years, we can see that the human and the machine to communication become more simple and direct and unimpeded is the future of human-computer interaction,and it is one of important developing direction of the research. The current scientific research is bacome one of the world hottest research topic..This thesis proposes a new nonlinear dimensionality reduction method for human gesture recognition and dynamic tracking based on a computer vision system. For the variation of posture appearance, the recognition and tracking of human hand gestures from one single camera remain a difficult problem. Here, we make some improvement to existing algorithm, get an unsupervised learning algorithm, distributed local linear embedding(DLLE) algorithm, to discover the intrinsic structure of the data. In the static recognition, First when the captured static image after preprocessing, then through the embedding algorithm of DLLE, these images are proposed to embedded in a low dimensional space, and pre-settings a database for the classification of the static gestures, which stored in the database is advance of the collection sample group. After that, use the probabilistic neural network (PNN) as a classifier, Similar input image will be classified in a group. In dynamic tracking,With a camera captured the gesture motions, use DLLE through the search to find the corresponding action position of the database, which make these captured images dynamically mapping to a low dimensional space. According to the classification of PNN, further calculated its gesture parameters and rebuild model. These capture samples as a gesture to the movement process sequence in the system, According to these gestures sequence of neighbor relations and the joint parameter model predefined in the database, we can calculate reconstruction joint parameters.Through the results of using the new algorithm gesture recognition and the analysis of LLE, NLE method with actual embedded in the comparison of the results, This new algorithm is proved that the significantly improvement in static recognition and dynamic tracking which obtained.