Research on Detection Algorithm of Hot Work Based on Target Recognition and Image Processing

Abstract

To enable real-time monitoring of fire operations within construction sites and to reduce the chance of fires, this paper proposes a detection algorithm that incorporates target recognition and image processing. Firstly, we add an attention mechanism to YOLO V7 to increase the localization accuracy of highlighted areas. Secondly, we use an image processing algorithm to identify the presence of fire in the localized area and recheck the localized highlighted area to improve the detection accuracy. After the verification is complete, this paper proposes a method based on monocular visual ranging to detect the presence of staff around the fire light to further confirm the authenticity of the fire operation and finally complete the detection of the fire moving operation. To verify the reliability and recognition accuracy of the algorithm, experiments were conducted on the dataset collected at the construction site, including ablation experiments on different target recognition networks and the application of different attention mechanisms. In addition to experiments on image segmentation of different highlighted areas located, and finally, extensive verification experiments on monocular visual ranging experiments are carried out. The experimental results show that the algorithm enables an effective detection of fire operations.