Moving object segmentation by background subtraction and temporal analysis

The problem of detecting moving objects is very important in many application contexts such as people detection and recognition, visual surveillance both in indoor and outdoor environments, and so on. In this paper we propose a motion detection algorithm based on background subtraction and shadow removing. The main idea is to implement a fast and reliable approach for motion detection, able to extract the moving objects without their own shadows, in a single-step algorithm. It is based on the correlation between regions selected from the reference image and the current one. In addition, the proposed algorithm is able to efficiently update the reference model: unlike traditional background updating algorithms, our approach works well on every pixel in the background image, even if covered by a foreground object, in order to have always a consistent reference image. The experiments, performed on real image sequences acquired both in indoor and outdoor environments with natural and artificial lights, demonstrate the effectiveness of the proposed algorithm