Moving Shadow Detection from Background Image and Deep Learning

摘要

We present a novel approach for moving shadow detection, which is applicable to various environments. Although there have been extensive studies of shadow detection since 1980s, the problem is still considered as a challenging and important issue in the most visual surveillance systems. Herein, we propose a shadow region learning method using a deep structure for moving shadow detection. Unlike previous approaches which are usually based on hand-crafted features using chromacity or physical properties of shadow regions, our approach is able to automatically learn features of shadow region from input source and its background image. The proposed approach is relatively simpler to implement than previous approaches as we don't need to consider intensity and color properties of video sequences. However, its performance is comparable to that of state-of-the-art approaches. Our algorithm is applied to five different datasets of moving shadow detection for comprehensive experiments.