VISUAL TRACKING-BASED HAND GESTURE RECOGNITION USING BACKPROPAGATION NEURAL NETWORK

摘要

This article proposed a hand gesture recognition using backpropagation neural network based on visual tracking. The proposed algorithm is passed through three stages, namely, hand region extraction, hand feature extraction and hand gesture recognition. The hand region is extracted from the background using skin color detection based on YCbCr color filter. The extracted hand region is then converted to gray scale and binary image to speed up processing time. The hand feature is obtained from the binary image of the detected hand region by dividing the feature into six regions. The last stage is the recognition of the hand gesture which is performed based on the backpropagation neural network. The six regions of the hand feature of each gesture become the input data of the neural network. In order to accommodate the six regions of the hand feature, the proposed method implemented six nodes on the input layer. Number of the hidden layers is two with ten nodes on each. The output layer has four nodes to accommodate ten output of the recognitions process. The hand gestures to be recognized are a set of ten hand gestures, namely: Stop, One, Two, Three, Four, Hello, Yes, No, OK and Call. The experiments on each hand gesture showed that our proposed algorithm can reach the good performance of recognition rate with minimum result of 80%, maximum result of 94% and average result of 86.67%.