Efficient Head Detection Based On Direction Gradient Histogram And Its Application In The Intelligent Monitoring System

Video surveillance system is playing a crucial role in military, economics and people’s daily lives. Traditional video surveillance is mainly conducted manually, which is quite insufficient and easily leads to careless omission. With the rapid research development of Computer Vision and Pattern Recognition, as well as the skyrocketing hardware performance, it is a general trend to utilize automated intelligent surveillance system. And it is imminent on how to detect human and get clear surveillance snapshots efficiently and quickly.Firstly, we conduct an extensive study of the state-of-the-art intelligent surveillance systems and person detection techniques. We also analyze their scope of application and limitations.Secondly, we give a performance analysis of the traditional histogram of oriented gra-dients head detection algorithm, propose a GPU accelerated algorithm which can achieve effective detection. We also conduct a statistical study of the high false positive rate that the algorithm might have, and ultilize a filtering algorithm based on motion and appearance which tremendously reduces the high false positive rate of the original algorithm.Finally, based on the requirement analysis of the video based human detection and clear image acquisition system, we give the design of our system. Which, in order to build the system with high efficiency, we adopt the MVC model to design and implement the system software modularly. Powered by adaboost face detection, histogram of oriented gradients head detection and differential EMD tracking, we implement the whole system which can detect, track human and get clear images in real-time, and also supports indexed storage of images that facilitates searching and browsing.