A hierarchical moving shadow detection method based on sparse representation in surveillance traffic video

摘要

This paper presents a novel shadow detection method based on Sparse Representation in intelligent traffic surveillance videos. First, the foreground is obtained by using Gaussian Mixture Model, and shadow detection is divided into rough and refined detection stage. Second, in the rough detection stage, we construct shadow dictionary and object dictionary by using training samples, the reconstruction error of foreground regions is adopted to determine whether the region is shadow or not. Meanwhile, brightness feature is utilized to separate shadows from foreground regions. In the refined detection stage, features derived from c_1c_2c_3 color space are applied to refine the rough detection result The final detection result is obtained after space adjustment. The proposed method is implemented on four databases, and compared with some well-known methods. Experimental results show that our method not only could detect the moving shadow accurately in traffic surveillance video scenes, but also is superior to several existing methods.