Research Of Hand Detection Method Based On Skin Color And Real AdaBoost

Hand Detection is a process that detects the human hands form videos or images. It is a key point of Hand recognition, Sign language recognition and Human detection. At the same time, it has broad prospects in the fields like Human-computer Interaction, Hand Recognition and Game Control.Hand Dection is a challenging task, because the hand is a complex articulated object that consists of many connected parts and joints. Based on the study on Hand Detection, the main research work involved in the following two aspects:A method based on Real AdaBoost for human hand detection is proposed in this paper. Real AdaBoost algorithm is improved on AdaBoost. It can obtain strong learners by combining a series of weak learners. It extended discrete weak classifiers to real-valued confidence-rated classifiers. Through analysis and the experiments, Real AdaBoost has a higher classification performance. In this paper, the cascade Real AdaBoost is obtained by training and used for testing.A hand detection method combined skin color segmentation and Real AdaBoost is proposed in this paper. Firstly, Skin color is the important information of the hand, and does not depend on the position of the hand and the size. It has rotation invariance and can distinguish other colors of most objects. The method based on Skin color segmentation is simple and fast, meanwhile, the robustness is strong. Before skin color segmentation, it must compensate the picture’s light because skin color segmentation is easily influenced by lighting conditions. Then, Skin color model in the YCbCr color space is built to segment the skin color reigns which include candidate human hand regions from the image. This process lays a foundation for the following step. Secondly, the candidate region is detected by cascade Real AdaBoost again, and finally the hand is detected. The method takes advantages of both. The new method is able to detect hands with low false negative rate and high detection rate. It can achieve more accurate hand detection.