Research on multi-camera data fusion for improving fire detection accuracy

With the rapid urbanization in China, the use of various electrical equipment and a large number of flammable materials has led to an increasing trend in the frequency of fires from year to year. In this paper, we start with data fusion to collect fire open data fragments so as to establish a fire detection dataset. A fire monitoring terminal that utilizes multi-feature fusion is created using the data fusion algorithm of the convolutional neural network to improve the main structure of the YOLOv5 fire detection model. The detection effect of the improved model is compared with other network models when combined. In this paper, it is found that the improved YOLOv5 model has better training time and steady state of training effect than the other three groups of models, and its mAP value is improved by 22.1%, 13.6% and 10.13% compared with the other three models, respectively. The average detection accuracy of the improved model for flames and smoke generated with different materials is also higher than that of the other three groups of models. At the same time, the improved model has stronger network classification and checking abilities, and is more accurate in recognizing whether a fire is occurring in the image. In this paper, by improving the YOLOv5 model, it is effectively applied to the fire detection work, realizing the dynamic analysis of real-time detection of flame and smoke and providing an effective detection model for fire monitoring. © 2024 Wen Wang et al., published by Sciendo.