HOG-like Feature for Hand Detection and Tracking

Abstract

In this paper we present a robust and real-time hand detection and tracking method. A new edge feature descriptor (HOG-Like feature descriptor) is first proposed, and used for the hand detection and tracking. The hand detection method is based on the AdaBoost and HOG-Like feature. A cascade framework is used to speed up. A detailed study was developed to select the training parameters. The HOG-Like feature can also be used for other object detection, such as president detection. When tracking the hand a six parameters-state and motion model is used for predicting the hand's location. In this way it reduces the computer time radically. The experimental result demonstrates that the method can successfully detect the hand and track it in real-time. The hand detection and tracing can be used hand-based Human-Computer Interface.