Spatial Feature Based Shadow Detection in Visual Traffic Surveillance System

Abstract

A shadow detection algorithm base on spatial features was proposed, in the case of focusing on traffic vehicle detection system. First of all, multi-foreground rectangles were extracted by using Gaussian mixture model (GMM) and edge detection operator of mathematical morphology. Then, histogram of horizontal location - foreground point number of vertical direction was computed, combined with optimum threshold segmentation, shadow areas were removed. To enhance the adaptability, the system learns direction relations between target and its shadow, automatically. Results were presented for several video sequences representing a variety of illumination conditions. Experimental results in different traffic conditions showed our technique robust, self-adaptive, and real-time.