Spotting recognition of human gestures from time-varying images
Takuma Nishimura, R. Oka · 2002
We study spotting recognition of human gestures from time-varying images. We propose a feature extraction method and a spotting method. The feature extraction reduces each frame of motion into a frame image of small size, such as 3/spl times/3 pixels. We show that the feature with 3/spl times/3 is the best one to provide robust characteristics to the changes of cloth and background. The spotting recognition rate was about 80% for 8 gesture categories. A new spotting method called non-monotonic continuous DP is proposed for spotting gestures and their variations such as reverse, partial and stop motions. We show the effectiveness of the new spotting method.