Improve shadow detection and removal in traffic surveillance system

Abstract

In Traffic Surveillance System, identifying the moving vehicles is one of the most basic function. However, when observing in sunshine conditions, shadows from vehicles, trees, house… always make vehicles detection difficult. Many methods have been done on vehicles shadow detection. Background subtraction methods unable to detect between shadows and vehicles because the shadow has similar dynamics to the object that it is cast by. Another methods incapability distinguish when vehicles have a dark color similar to the shadow. Furthermore, it can cause some problem such as object merging and shape distortion since they typically differ significantly from the background. In this thesis, I will develop a new method by improving moving shadow detection and removal methods through properties of shadow. This method can achieve results that are more accurate and operate well in the traffic condition in Vietnam especially in Ho Chi Minh City. Keywords: Shadow detection, Lab color, contour, Canny filter, morphology, quantitative performance, surveillance system.