基于静态特征和动态行为的火灾检测模型

Fire detection model based on static characteristics and dynamic behavior

为实时准确地检测视频监控的火灾信号,减少火灾误报,提出1种基于静态特征和动态行为的火灾检测方法。利用改进的Faster RCNN检测模型,根据可疑火灾区域的颜色特征和空间特征对其进行目标检测和特征降维,与传统的Faster RCNN相比平均检测精度提升5%;利用ILSTM对连续帧中的特征进行累加,对短期内是否发生过火灾进行分类。连续的短期决策在1min内以多数票表决最后决策。结果表明:方法将火灾检测的准确率提升到97.92%,并成功解释火焰和烟雾的时间行为。

In order to accurately detect the fire signals of video monitoring in real-time and reduce the false alarm of fire,a fire detection method based on the static characteristics and dynamic behavior was proposed.Firstly,the improved Faster RCNN detection model was used to carry out the target detection and characteristics dimension reduction according to the color and spatial characteristics of suspicious fire area.Compared with the traditional Faster RCNN,the average detection accuracy was improved by 5%.Then,the ILSTM was used to accumulate the characteristics of continuous frames,and classify whether there was a fire in the short term.Then,the continuous short-term decision-making was decided by a majority vote within 1 min.The results showed that the accuracy of fire detection was improved to 97.92%by this method,and the time behavior of fire and smoke was explained successfully.