Complex Background Of The Head And Shoulders Like The Detection And Location

Head and shoulder detection is one of the important problems on human body analysis. It is the first step of face detection and the base of pedestrian detection. It can be applied to Content-based Image Indexing and Smart Surveillance Systems. It is essential to develop robust and efficient algorithms to detect head and shoulders.In this thesis, we present an algorithm based on Support Vector Machine (SVM) and Discrete Wavelet Transform (DWT) to detect upper body of human in gray images under complex background. This algorithm is made of training section and detecting section. A lot of head-and-shoulder samples and "not head-and-shoulder" samples are collected as training examples. Features are extracted from wavelet coefficients of these training images, and used to train a combined classifier which consists of a linear SVM classifier and a nonlinear SVM classifier. The detection system works by testing whether there is a head and shoulder in each window of the image.The thesis gives the experiment result of our algorithm. The result shows that the algorithm is able to obtain accuracy of 87. 58% over a test set of 85 images containing 153 targets.We use this algorithm to detect the motion images and adopt the background separate technique to get the area which has the upper body of human, and then adopt SVM method, which has very high detection rate, to detect in this area. Because it only detects this small area, this method can be considered as the foundation of real time human motion detection as well as guarantees the accuracy of detection.