Research On Finding Hand Position In Complex Background

Hand detection is based on machine vision and image processing. People's intention can be intuitivly known by gestures. Unfortunately, hands have a flexible shape, for example, hands have different angles of finger joints and hands sometimes may be partially occluded, which can lead to a certain degree of bad influence in practical application. So it is a challenging task to detect and locate hands accurately in still images. On the basis of studying the knowledge of hand position in the gesture detection and identification, a series of technical problems on feature extraction and positioning methods are studied to design a reliable hand positioning system. A hand location method based on cascade classifiers is proposed in this paper.1. The preparation of a hand database. Firstly several feature points are selected manually in the hand images to represent the position of the human hand in the images, procrustes algorithm is used to adjust the collected hand images to standard position. The images are cropped and have a certain degree of rotation, translation and scaling.2. The Histogram of Oriented Gradients(HoG) feature is used in this paper. AdaBoost algorithm and Multi-Branch Tree algorithm are used on training of hand detection classifiers. The first stage of AdaBoost algorithm is trained for fast excluding non-target by simple features, cascade structure is used to train strong classifiers. A method is proposed by calculating average tree threshold to achieve higher accuracy in Multi-Branch Tree algorithm. The experimental results show that the proposed methods can effectively complete the positioning function on detecting human hands.