The current fire detection methods still have the problems of low detection accuracy and high computational resource consumption.This paper designs a highway fire detection method based on the motion of smoke combined with deep network.The moving foreground in the video is extracted by Gaussian mixture background modeling,and the region of interest is select-ed for the moving foreground to obtain the potential smoke area,and the HSV color space analysis is performed on the area to determine the presence of smoke.For video frames with smoke,a highway-specific fire and smoke dataset with more than 50000 object annotation boxes is constructed,combined with the YOLOv5 detection method,to achieve a mean average precision(mAP)of 90.16％.The proposed method avoids frame-by-frame fire detection,greatly reduces the waste of computing re-sources,and has research and practical engineering application values.
关键词
烟雾检测/火灾烟雾数据集/交通视频分析/深度学习/高速公路火灾
Key words
smoke detection/fire and smoke dataset/traffic video analysis/deep learning/highway fire
维普
万方数据