Real-time hand gesture detection and recognition for human computer interaction

摘要: This thesis focuses on bare hand gesture recognition by proposing a new architecture to solve the problem of real-time vision-based detection, tracking, and for interaction with an application via gestures. The first stage our system allows detecting tracking in cluttered background using face subtraction, skin detection contour comparison. second recognizing gestures bag-of-features multi-class Support Vector Machine (SVM) algorithms. Finally, grammar has been developed generate commands control. Our consists two steps: offline training online testing. In stage, after extracting keypoints every image Scale Invariance Feature Transform (SIFT), vector quantization technique will map from into unified dimensional histogram (bag-of-words) K-means clustering. is treated as input SVM build classifier. testing frame captured webcam, detected my algorithm. Then, are extracted small that contains posture fed cluster model them bag-of-words vector, which classifier recognize gesture. Another was proposed Principle Components Analysis (PCA). most eigenvectors weights images determined. projected onto form its test weights. minimum Euclidean distance determined among each Two gesture-based 3D gaming virtual environment were implemented. exertion videogame makes use stationary bicycle one main inputs game playing. user can control direct left-right movement shooting actions set commands, while game, helicopter over city commands.