视频图像火灾烟雾识别仿真研究

Video Wildfire Smoke Detection Algorithm Simulation

研究视频火灾烟雾的准确识别问题。烟雾在有强风干扰的情况下会丢失向上飘动特征,并且烟雾浓度和运动速度发生剧烈变化使火灾图像不在确定范围之内。传统烟雾识别方法多是在光流法的基础上指定向上飘动的主运动方向和烟雾运动速度的范围进行检测,会造成光流特征失效,导致识别率不高。为解决上述问题,提出了一种采用光流特征的烟雾识别算法,首先通过运动和颜色检测提取疑似烟雾区域,然后运用Horn-Schunck(HS)光流法得到像素的运动速度和方向,进而提取光流速度及方向的均值和方差、光流对比度和方向一致性作为特征,最后将光流特征组成的特征向量作为支持向量机(SVM)的输入,进而利用构建的二类分类器对烟雾进行识别。实验结果表明,改进算法能有效识别火灾烟雾,具有较强的抗干扰能力和鲁棒性。

A new smoke detection method is proposed based on optical flow. Firstly,dubious smoke areas are seg- mented by the motion and chrominance detection, then the Horn-Schunck optical flow algorithm is used to obtain pix- el motion vectors. Further, the averages and variations of the optical flow velocity and the orientation, the contrast of optical flow and the orientation consistency are extracted as eigenvalues. The eigenvalues are used as the input vectors of SVM, then a 2-class classifier constructed to detect smoke. The experimental results show that the proposed meth- od can detect wildfire smoke effectively and has strong anti-interference ability and robustness.