The Research Of Automatic Statistical Technology For Pedestrian Flow Based On The Image Processing

With the development of society, the crowd in various kinds of public places flows more and more frequently. How to manage and control the crowd effectively comes to be an important issue which need to be resolved urgently. With this background, intelligentized surveillance crowd technology arises at the very moment and its application foreground becomes wider and wider.This paper discusses the automatic statistical technology of pedestrian flow in video sequences gathered from a fixed camera and completes the system design with Visual C++ 6.0 software development platform based on windows operating system.The main research contents of the paper include motion detection, head detection and object tracking. In the research of the motion detection, traditional single-Gauss background model is improved inthe steps of establishment and update of background in this paper. With the improved algorithm, moving objects can be segmented from the image effectively. In the post-processing, Unger algorithm and Blob detection algorithm based on fan-shaped are applied to the system to remove glitches and noise in the image quickly without destructing the contours of moving objects.In the research of the head detection, a novel algorithm which can be better used to detect head from moving human are proposed focusing on the profile characteristic and color feather of person's head. This algorithm can be used in sparse crowd where human heads are not obstructed seriously.In the research of the object tracking, human's head is selected as tracking target and the center of the ellipse which is fitting human's head is used to be the track point. It can reduce the computational complexity. A novel object tracking algorithm based on Kalman filter which can be used to track heads in video sequences are proposed on the basis of common object searching and tracking algorithms.Finally, the research work of this dissertation is summarized and some existing insufficiencies in the research results are pointed out .This paper also makes some suggestions for further research on the realization of the system and the theories.