Intersection foreground detection based on the cyber-physical systems

Foreground detection is an important and challenging task in the traffic surveillance applications, especially at urban intersections. A very common solution for foreground detection with static cameras is background subtraction. When the background is modeled, learning rate is an important parameter and is difficult to select. Difficulties of background updating usually come from complex physical scenarios such as sudden illumination changes and occlusion. Conventional background modeling algorithms usually use the same learning rate for the entire frame or sequence and are not able to match the environment changes adaptively. Although some improvements with adaptive learning rate have been achieved, they are still not appropriate for complex traffic intersection scene. This paper addresses the problem of background subtraction at busy intersection and proposes a new method for vision-based foreground detection. It assigns a learning rate adaptively for each pixel according to the information collected by the infrastructure-based networked system, also called a Cyber-Physical System (CPS). Our goal is to provide a simple and efficient solution to improve the robustness and accuracy of intersection foreground detection. We test our approach in urban traffic intersection and the experimental results show that the new method has a promising future.