Vision-based illegal human ladder climbing action recognition in substation

Abstract

Nowadays, unattended monitoring system has been widely used in substation for its efficiency and efficacy, and it sometimes may cause safety problems for utility workers. In order to ensure workers' safety, in this paper, we focus on the problem of illegal human ladder climbing action recognition in substation using vision-based algorithm. Specifically, we first detect “forbidden” and “allowing” types of signboards on the ladder to localize the ladder and then define the unsafe area. We use HSV-based algorithm and do hough circle detection to recognize “forbidden” signboard. To recognize “allowing” signboard, we propose HOG-based feature extraction algorithm with SVM classifier, and then use color analysis for further detection. After that, we detect and localize human action applying Visual Background Extractor(ViBE) algorithm. Finally, we can recognize illegal human ladder climbing action based on the relative position between human and signboards. The experiments demonstrate the relative high accuracy of our proposed algorithm.