Abstract
This paper presents a learning-based solution to tackle the real-time gesture recognition of bimanual (two hands) gestures which is not well studied from the literature. To overcome the critical issue of hand-hand self-occlusion problem common in bimanual gestures, multiple cameras from diversified views are used. A tailored multi-camera system is constructed to acquire multi-views bimanual gesture data, and data from each view is then fed into a separate classifier for learning. Thus, to ensemble results from these classifiers, we proposed a weighted sum fusion scheme of results from different classifiers. The weightings are optimized according to how well the recognition performed of the particular view. Our experiments show multiple-view results outperform single-view results.