A comparative study of different color spaces for foreground and shadow detection for traffic monitoring system

Abstract

Segmenting foreground objects of interest in real time is an important step in many applications of video surveillance, vehicle tracking and traffic monitoring. Background subtraction is a method very often used to segment moving objects in image sequences. In this paper we present a fundamental unbiased study of different color spaces for detecting foreground objects and their shadows in image sequences. Different color spaces show different efficiency in detection of foreground objects and their shadows. This empirical study was done with the motivation of determining, which color space is best for foreground segmentation and shadow detection. This study of the quality of foreground and shadow detection as a function of color space is unique kind and especially relevant to color image sequences. Our study includes five color spaces RGB, XYZ. YC/sub r/C/sub b/, HSV and the normalized rgb. We use an empirically substantiated model of shadows formulate the detection scheme for each color space. We use a statistical technique to model the background pixels. The results are compared in terms of true detection, misses and false detection of pixels and also detection of the moving foreground objects as blobs. The results show that YC/sub r/C/sub b/ is the best color space for optimal foreground and shadow detection.