Real Time Trajectory Based Hand Gesture Recognition

The recognition of hand gestures from image sequences is an important and challenging problem. This paper presents a robust solution to track and recognize a list of hand gestures from their trajectory. The main tools of the proposed solution are robust kernel density estimation and the related mean shift algorithm, used in both video tracking and trajectory segmentation. The gesture definition is based on strokes in order to allow the use of a low complexity gesture recognition method. The gesture recognition process is trivial, being reduced to a syntactic analysis of the feature vector avoiding the necessity of complex classification methods based on curve matching. Despite the restrictions derived from the stroke based definition of gestures, the low computational complexity of the algorithm allows its implementation on low-cost processing systems. Key-Words: gesture recognition, mean shift, robust methods, video tracking, human machine interface.