微型电脑应用

隧道火灾会引起交通中断，严重时会危及人员、财产与隧道结构的安全。目前的火灾检测算法不能对隧道监控视野中存在车灯闪烁等干扰情况时进行有效检测，对此，提出一个隧道火灾视频目标检测数据集，并采用联合多帧检测的YOLOV网络得到火灾检测结果，对结果进行火焰烟雾面积变化率的统计，降低火灾的误报，提升检测结果的可靠性。在多种隧道实际火灾视频中的测试表明，所提方法达到了90.53%的平均检测精度与26.75 f/s的检测速度，能够识别距离摄像机较远的火焰、区分多种类别灯光的干扰，具有较高的检测可靠性与实际应用价值。

Tunnel fires cause traffic interruption, and in severe cases may endanger the safety of personnel, property and tunnel structure. The current fire detection algorithms cannot effectively detect the interfence such as flickering lights in the tunnel monitoring perspective. This paper proposes a tunnel fire video object detection dataset, and uses the YOLOV network that combines multiple frames for object detection to obtain the fire detection results, and performs statistics on the change rate of the flame smoke area to reduce the false alarm of fire and improve the detection results reliability. Experiments on actual fire videos in various tunnels show that the proposed method achieves average detection accuracy of 90.53% and detection speed of 26.75 f/s. The proposed method can identify flames that are far away from the camera, distinguish multiple types of lights and other distractors, and has higher detection reliability and practical application values.