Experiments with computer vision methods for hand detection
Zhong Zhang, Rommel Alonzo, Vassilis Athitsos · 2011
For gesture and sign language recognition, hand shape and hand motion are the primary sources of information that differentiate one sign from another. So, building an efficient and reliable hand detector is an important step for recognizing signs and gesture. In this paper we evaluate four features for hand detection: color, temporal motion, gradient norm, and motion residue, and we explore the potential of these features for building a reliable hand detector. At first, we use these four features separately to identify where the hands are in each frame of our gesture videos. Then we evaluate different combinations of such features using weighted linear combination, so to build a more accurate hand detector. Experimental results show the relative performance of the four features in isolation and in different combinations, and demonstrate promising results for detectors that combine these features.