Research On Pedestrian Detection Using Background Subtraction And Simple Behaviors Recognition

With a rapid development of the economy and the society, the demands of security alarm system used in the area of bank, power and transport increase rapidly. And there are higher requirements of the intelligent video surveillance technology. Pedestrian detection and behavior recognition system become a major research topic in the intelligent video surveillance technology. A typical system of the pedestrian detection and recognition include such modules as pedestrian detection, shadow removal, human tracking and behavior recognition, making a good sense of both helping security to identify and deal with abnormal behavior and unmanned surveillance all day. The pedestrian detection, shadow removal and behavior recognition have been researched in this paper.Motion detection is the basis for pedestrian detection and behavior recognition technology. After a deeply research of the frame difference and background subtraction method used in the moving detection, a new module call CW4 based on background subtraction has been proposed in this paper. The algorithm combines advantages of the frame difference and background subtraction method and makes a good use of brightness, hue and color information. Better accuracy has been achieved in the premise of real-time.Shadow removing has been studied to detect target more accurately. A model based on color calculation with weight and a model based on background divison have been proposed to remove the shadow in the color Image or gray image after a deeply research of the image's pixel and shape characteristics.A good performace has been achieved.The techniques on pedestrian posture recognition, pedestrian detection crossing the virtual lines and the number of person detection in a virtual region have been studied. A template-based posture recognition algorithm according to pedestrian features in binary images has been porposed. An algorithm used to judge whether the people cross the lines and an algorithm which can be used to count the number of the persons who are in the specified area according to the relationship between the person's location and the virtual line have been proposed. The algorithm of pedestrian detection, shadow removal and behavior recognition will play an important practical and theoretical significance in the fileds of intelligent security monitoring, intelligent transportation and visual navigation.