Detecting Moving Shadow using a Fusion of Local Binary Pattern and Gabor Features

Abstract

Detecting moving objects in the video sequences is an important issue for visual surveillance. It is needed to extract the moving shadows to describe the moving targets correctly. This paper proposes a method for detecting shadow by exploiting the similarity of texture between the segmented regions of moving objects and the corresponding regions in the background image. The proposed method generates texture similarity maps using Local Binary Pattern (LBP) and Gabor features. Principal Component Analysis (PCA) fuses the feature maps to generate an integrated texture similarity map to classify shadow pixels. HSV color model is then exploited for validation of the shadow pixels and reduce the misclassifications of both shadow and moving objects pixels. In the end, post processing step of morphological operations to determine the final shadow regions. Experiments show the effectiveness of the proposed method in shadow detection process and provide high-performance rates.