基于光流改进与YOLOv3的烟雾检测方法

Smoke detection method based on optical flow improvement and YOLOv3

火灾严重威胁着人们的生命和财产安全,而烟雾作为火灾前期的一个重要特征,应当作为火灾检测的首选目标。为了解决传统火灾检测在准确度和实时性上存在的不足,提出了一种基于传统光流改进与YOLOv3结合的烟雾检测模型。该模型针对烟雾的动态特征,通过改进光流算法对动态前景区域进行目标框定完成一次筛选;利用大样本数据集训练的YOLOv3网络模型,将初筛结果输入模型进行二次识别和筛选,从而达到检测烟雾目的。实验结果表明:在各类烟雾检测任务中,该模型可以有效地减少外界因素干扰,准确实时地检测各类烟雾情景,鲁棒性较好。

Fire is an important factor threatening people’s life and property safety,and smoke,as an important feature in the early stage of fire,should be the first target of fire detection.In order to solve the shortcomings of traditional fire detection in accuracy and real-time,a smoke detection model based on traditional optical flow improvement and YOLOv3 is proposed.Aiming at the dynamic characteristics of smoke,the improved optical flow algorithm is used to filter the target frame from dynamic foreground area for the first time.Then the YOLOv3 network model trained by large sample data set is used to recognize and filter the smoke target with the first filtered results as input into the model for the second time.So the purpose of smoke detection can be achieved.The experimental results show that in all kinds of smoke detection tasks,the proposed model can effectively reduce the interference of external factors,accurately detect all kinds of smoke scenarios in real time with good robustness.