Study Of Human Motion Detection And Tracking For Vehicle Driver Assistance Systems

Pedestrian detection technology is an active safety technique which guarantees pedestrians safety. It has become a hot research topic. especially in the urban traffic environment. Pedestrian detection technology can warn the driver that the vehicle is possibly colliding with the adjacent barriers, it provides a strong technical support that can avoid traffic accidents and obvious social and economic benefits can be received.The paper researches on the human motion detection and tracking technology in the vehicle driver assistance systems. In the case of static camera, detects the moving object, identifies and tracks the pedestrian in the video image captured. Moving object detection is a process of extracting moving object from a sequence of images, it is the foundation of the follow-up treatment and is of great significance. This paper analyzes three kinds of common moving object detection algorithm (Background Subtraction method, Consecutive Frames Subtraction method, Optical Flow method), integrates the merits of the Background Subtraction method and the Consecutive Frames Subtraction method, using an improved Consecutive Frames Subtraction method to update the background model to obtain the background image, then combines with the Background Subtraction method to detect the moving objects.As to discern whether the objects detected are human, the paper quickly distinguishes human from other objects based on the shape and velocity characteristics of pedestrian, in this way it can determine whether the object detected is the moving pedestrian. In the tracking phase, the paper establishes a Kalman filter model to track pedestrians, chooses the location and tracking window of pedestrians as the state variables, and sets up the state equation and observation equation to track pedestrians effectively in real time. In the case of dynamic camera, the paper uses a block-matching method to achieve the global motion estimation, thus removes the impact of the background movement, and then accurately detects the moving objects in the dynamic context combining with the Consecutive Frames Subtraction method. In this paper, the form of hand-held camera is adopted to simulate the can zoom camera loaded on vehicle, and makes experiments with the video images captured, the experiment results show that the algorithm proposed can detect, identify and track pedestrians, and has a certain robustness.