A study on hand motion recognition using tracking with active cameras and template matching

摘要

In this study, we propose a hand motion recognition method for a human-computer interface using active camera tracking and template matching. Most of the conventional methods use features extracted by image processing for recognition, such as the fingers and the contour of a user's hand. However it's difficult to extract these features stably using a few cameras because of self-occlusion. Therefore we treat an input image itself as a feature vector and recognize hand motion by referring it to the shape dictionaries, which store a large number of motion templates. We experimented by using shape dictionaries constructed from sets of sampled patterns. The result showed that our proposal method could recognize hand motion satisfactorily in real-time.