火灾烟雾的轮廓光流矢量分析

Contour Optical Flow Vector Analysis of Fire Smoke

由于烟雾运动具有特殊的湍流特性,使得其运动方向具有特定一致性.利用这一特性,采用光流矢量分析方法,对可疑区域运动方向进行判定.为了提高算法实时性,首先提取可疑区域轮廓,然后对轮廓光流矢量进行分析.通过对比烟雾与非烟雾干扰（包括运动的光照、行进中的汽车车灯、水蒸气等）,分析其轮廓的光流矢量特性,发现烟雾轮廓与非烟雾轮廓的光流矢量具有明显差异性,即其运动矢量基本呈现向上、向外扩散的趋势,因此角度基本位于一、二象限,即在0°~180°之间（除个别干扰点）,这也与烟雾湍流运动特征相匹配,而分析非烟干扰发现,运动矢量并不具备这一特征.通过对比实验证明了该算法在保证准确度的基础上,将传统光流算法的计算速度提高了3倍.

The moving directions of smoke have a certain consistency because of its special turbulent characteristics,which is used to judge the movement directions of suspicious area based on optical flow vector analysis.In order to improve the real-time performance of the algorithm,the contour of suspicious area was extracted firstly,then the optical flow vectors of contour were analyzed.Comparative experiments of smoke and non-smoke disturbance（including moving lights,traveling car headlights,steam,etc）have proved that the optical flow vectors of smoke and non-smoke outline have obvious differences.The moving vectors of smoke present the trend of outward and upward diffusion. Therefore,the angles of vectors are basically located in the first and second quadrants,that is,between 0 °and 180 °（except very few disturbance points）.This result also matches the turbulent characteristics of smoke.However,the moving vectors of non-smoke disturbance do not possess this feature.The experimental results show that the calculation speed of the proposed algorithm is increased by three times compared with the traditional optical flow algorithm on the basis of ensuring accuracy.