Vision based two-level hand tracking system for dynamic hand gestures in indoor environment

Object detection and tracking is one of the most challenging phases in any image processing application. The challenges that arise during hand gesture recognition are: hand shape variation, presence of multiple persons, illumination variation or occlusion with background. In this paper, a robust hand detector and tracker algorithm has been proposed which overcomes the above difficulties to a great extent. The proposed system first detects hand using a combination of color and motion information. Then, the hand is tracked using combination of feature based tracker modified KLT algorithm and a color-based tracker Camshift algorithm. New feature points have been added to our modified combination of feature and color based tracker. This prevents the loss of tracking as seen in traditional KLT algorithm. The experiments carried out using the proposed system shows that the system provides better results in comparison to the existing literatures on various challenging hand sequences.