Skeleton Modeling And Hand Gesture Recognition In3D Human Computer Interaction

Human-computer interaction is becoming an important part in people’s daily life.How to interact between human and computer more effectively is becoming an hot topicin computer science. Being more widerly used, it provides the user with more naturaland more pretty interaction mode, with the input of human hand gesture.Human-computer interaction and gesture recognition is widely used in remote control,portable media player, XBOX game player, portable computer, and so on.Hand gesture recognition includes dynamic hand gesture recognition and handposture recognition. People can interact with computer using only a series of movementor one hand posture. In hand gesture recognition, the most critical algorithm is thelocalization and tracking of joints in human body and hand. Computer can understandhand gesture if the3D skeleton model or hand skeleton model is established. Therefore,this thesis puts forward new algorithm for3D human skeleton extraction and handmodeling. Then, it puts forward new algorithm of hand gesture recognition usingD2LS(Direction to Location Similarity) and multi-layer classifier, with very goodresults.The main contribution of this thesis is in the following:It puts forward new algorithm for high-speed3D skelton extraction using centerlines. It also puts forward3D human body part recognition algorithm, using which toprecisely establish3D skeleton, with high-speed and precise results. And then, thisthesis puts forward new3D skeleton tracking algorithm, integrating detection andtracking steps into one advanced system, providing good tracking results. This thesisputs forward the algorithm of corner detection and matching for tracking, settling theproblem of tracking objects of light-chaning and shape-changing. New algorithm istested on PRMI-Skeleton-21dataset, which contains large-scale of pictures. Newalgorithm has very good results.This thesis puts forward new algorithm of multi-layer classifier using stroke-basedfeature, and then use D2LS (Direction to Location Similarity) theory to recognize hand gesture. New algorithm runs with high speed and high accuracy in hand gesturerecognition. Then, this thesis puts forward new hand detection and tracking algorithmusing HBPD. New algorithm has much better performance than the traditional ones.Gesture-16dataset is used for testing, indicating that new hand gesture recognition, newhand detection and new hand tracking algorithm are much better than the traditionalones, with higher speed, and more precise results.This thesis puts forward new hand modeling algorithm, establishing the whole3Dhand skeleton model, which contains all joints location. In the new3D hand skeletonmodel, fingers can be curved. The whole3D hand skeleton model has very good results.