Research on transformer fire flame detection and localization technology based on Binocular Vision

Transformer fire is hidden, three-dimensional, and multi-scale, which is difficult to detect. In this paper, combined with the efficiency of deep learning in flame detection and the positioning advantages of binocular stereo vision, the transformer fire flame detection and positioning algorithm is proposed, and verified by experiments. The experimental platform of transformer oil pool fire was built, 122 sets of positioning experiments were carried out with 6 kinds of camera baselines, and the positioning rules and ranging range of the binocular system were determined. Under the condition that the camera baseline is 50 cm, the transformer fire flame detection and positioning experiment is carried out. The results show that the positioning error at 3 m to 8 m can be kept within 0.2 m, and the positioning error at 6 m is 1.3%.