An Effective Moving Cast Shadow Detection Framework for Surveillance Video Analysis
Jiangyan dai,Huihui Zhang,Jin Gao,Yugen Yi,Bingshu Wang
DOI: https://doi.org/10.2139/ssrn.4170609
2022-01-01
SSRN Electronic Journal
Abstract:Moving object detection is crucial in intelligent surveillance system, such as object tracking, object classification, and so on. However, moving cast shadows seriously degrade the accuracy of moving object detection and are often detected as objects, resulting in object merging, object loss or shape distortion. To overcome the drawbacks, we propose an original moving shadow detection approach by incorporating the broad learning system. Firstly, according to the color constancy and texture consistency of shadows, the color, texture and gradient features are extracted by input frames and a background. Then, the broad learning system is introduced, which can map input samples into a more appropriate feature space. Meanwhile, the BLS offers an alternative way, which makes it classify moving cast shadows effectively and efficiently. Finally, a post-processing is designed for keeping complete moving objects. Extensive experiments verify the proposed method’s effectiveness and generalization ability on outdoor and indoor surveillance scenes.