TRACKING NON RIGID OBJECT LIKE HUMAN HAND WITH CLUTTERED BACKGROUND USING CASCADE OBJECT DETECTOR

Abstract

This paper presents an algorithm for initialization and tracking of hand in a video taken by an ordinary color camera. Hand detection is carried out by training the Viola-Jones classifier using a histogram of gradient features. The detected hand instances contain false positives which are further filtered using skin color based segmentation. For skin color detection, theTint, Saturation, and Lightening (TSL) color space has been used, which is designed specifically for the skin color detection. As skin color is a perceptual and subjective human concept and not a physical property, the classifier has been trained on the skin and non-skin data with an AdaBoost algorithm, which provides the distinction between skin and non-skin input. A tracking algorithm is then used to track the identified hand in successive frames of video.