A New Bolts-Loosening Detection Method in High-Voltage Tower Based on Binocular Vision

• Published in: 2023 29th International Conference on Mechatronics and Machine Vision in Practice (M2VIP) pp. 1 - 6
• Main Authors: Wang, Qianshi, Zhang, Haodong, Lin, Chuang, Yang, Yang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 21.11.2023
• Subjects: 3D reconstruction; binocular camera; bolt; Fasteners; High-voltage techniques; neural network; Poles and towers; Robustness; Safety; Three-dimensional displays; Transforms; transmission towers
• DOI: 10.1109/M2VIP58386.2023.10413415

The loose connection of bolts on high-voltage transmission towers can adversely affect the regular operation and safety of the power system. Existing methods for bolt looseness detection suffered from inefficiency and missed detection. To address these issues and achieve high-precision real-time bolt looseness detection, this paper proposes a method based on binocular vision and neural networks. After capturing images of the bolts using a binocular camera, a Fast Region-based Convolutional Neural Network (RCNN) is utilized to locate the bolt connections. Then, adaptive threshold segmentation and Hough transform are employed to obtain the contours and corners of the hexagonal nuts. The obtained corners are used for stereo matching and 3D reconstruction, enabling the calculation of the distance between the nuts and the bolts ends to determine the looseness. The proposed method achieves a distance measurement error of 1.1%, an accuracy rate of 98.5% for looseness detection, and an average processing time of only 0.41s, thus realizing high-precision and fast detection.