| Real-Time Spotting Recognition of Gesture Motion Image and Spontaneous Speech (1994) | |||||||||||||
Abstract | |||||||||||||
| INTRODUCTION A spotting algorithm is proposed to recognize the meanings of 1) human gestures from motion images, 2) spontaneous speech. The spotting algorithm removes the need for temporal segmentation of gesture/utterance duration and introduces frame-wise recognition suitable for realizing a real-time system. We carried out some experiments into gesture/utterance recognition and the results have confirmed that: 1) for gesture motion images our algorithm is robust with various clothing textures and backgrounds, 2) for spontaneous speech our method allows speakers relaxed utterance. 2. SPOTTING RECOGNITION Let us show our matching algorithm between an input feature sequence and a set of standard sequence patterns stored in advance. This matching method is call "spotting recognition" and the "Continuous Dynamic method (CDP) [1, 2]" is one of the most famous of these methods. Let us define the feature vector at time t by<F34 | |||||||||||||
Publication details | |||||||||||||
| |||||||||||||