Study on Personnel Detection Based on Retinex and YOLOv4 in Building Fire

Abstract

When a fire occurs in a building, the internal environment is full of dense smoke, which will greatly hinder the evacuation and rescue of the trapped persons. If the evacuation and rescue are not in time, the life safety of the trapped persons will be seriously threatened. In response to this problem, this paper proposes a method for quickly detecting trapped persons in building fires. This method uses a combination of multi-scale Retinex image sharpening algorithm and YOLOv4 person detection algorithm. First obtain the image information of the fire scene, use the multi-scale Retinex algorithm based on the Gaussian pyramid to perform the sharpening process, and then use the YOLOv4 model to perform the personnel detection on the sharpened fire scene image. The experimental results show that the confidence of image person detection after Retinex sharpening processing has been significantly improved.