Improving Background Subtraction Algorithms with Shadow Detection

Motion detection of objects on video is a resource–intensive area of computer vision that requires several stages. The first stage of this task is the background subtraction process. The quality of further recognition fully depends on this stage. Natural or artificial illumination changes could provide a lot of shadows on the video. Such shadows could be falsely interpreted as parts of a moving object or even as the separate objects on the scene. This article proposes an improvement to the background subtraction algorithms by detecting and removing shadows from the images. The proposed approach relies onto spatial correlations between neighboring pixels in combination with the normal distribution of noise. It provides good background subtraction results along with low computational cost.