Fast safety distance warning framework for proximity detection based on oriented object detection and pinhole model. (15th March 2023)

• Record Type: Journal Article
• Authors: Li, Hao; Qiu, Junhui; Yu, Kailong; Yan, Kai; Li, Quanjing; Yang, Yang; Chang, Rong
• Is Part Of: Measurement. Volume 209(2023)
• Publication Date: 2023-03-15
• Subjects: Distance measurement -- Electrical safety -- Insulators -- Oriented object detection -- Pinhole model; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2023.112509
• Languages: English
• ISSNs: 0263-2241

Abstract

Abstract: Unauthorized approaching dangerous devices can cause serious dangers in electricity industry. Estimation on distances between human and these devices can effectively reduce the probabilities of various accidents, but there are limited studies focusing on it due to the complexity. In this paper, we propose a fast safety distance warning framework to detect proximity to dangerous devices in electrical operations. The framework consists of a customized oriented object detection model to extract precise pixel widths of objects, and a distance estimation method based on monocular camera and pinhole model to estimate distance. A tracking algorithm is applied to achieve targeted warning and fewer false alarms. The framework has been put into use in transformer stations in Yuxi power supply bureau, Yunnan Province, and distance estimation errors can be restricted to 0.5 meters. In experiments, the framework achieves 34 frames per second and 49.5% average precision, which are both state-of-the-art performances. Highlights: A computer vision-based solution for detecting approaching dangerous. A fast method based on monocular camera to estimate distances between objects. Using object detection and tracking to achieve a targeted warning mechanism. The method can be integrated in more monitoring systems for wider applications.