Shadow Detection Based on Combinations of Hessenberg Decomposition and Principal Component Analysis in Surveillance Applications

Abstract

ABSTRACTShadow removing is an important issue in performance of intelligent transportation systems. Moving shadows cause confusion and errors in moving object detection, since they cannot be removed from foreground using commercial background subtraction methods. In this paper, a synthetic shadow detection algorithm using combination of Hessenberg decomposition (HD) and principal component analysis (PCA) is proposed. The HD and PCA are applied to improve the performance of shadow detection. At first, the background image is obtained using moving average method and adaptive background estimation. The moving objects in the foreground image can be accurately obtained via the background subtraction method. In this algorithm, the candidate shadow region is estimated by using HD. With regard to ability of HD in the decomposition of foreground image into two parts as shadow and object, we should calculate the HD of foreground image and determine the shadow and object regions. In this procedure, some parts of vehicles like dark region and windshield that is similar to shadow are detected as moving shadow. Therefore, PCA is applied to compensate misclassified regions in the previous stage. The proposed algorithm is evaluated on real and operational conditions such as different perspectives and illuminations. Experimental results demonstrate the efficiency and effectiveness of our proposed algorithm in the intelligent transportation systems.