Abstract: Aiming at the problem of lower real time, higher false alarm rate and miss rate in the existing video fire smoke detection methods, the feature that early smoke movement is slow, main movement trend is upward and pixel intensity change is consistent in successive frames is found after depth analysis of smoke  image characteristics. Multiple feature smoke detection is realized through the Harris detection algorithm to find intensity changes and image edge feature point, based on the optical flow and the motion field correspondence by imaging plane optical flow estimation smoke changes in relative motion, calculation of motion vector information. This algorithm based on the smoke intensity change feature point as detection object greatly reduces the amount of data and shortens processing time. Because the smoke local and global characteristics are studied and applied, the proposed algorithm has the strong robustness and high detection accuracy rate.

Key words: Harris, feature point, Lucas-Kanade target tracking, smoke features, smoke recognition

摘要： 针对现有的视频火灾烟雾探测方法实时性差，误报率和漏报率都比较高的问题，在深入分析烟雾图像特征的基础上，发现早期烟雾运动缓慢且主要运动方向呈向上趋势，在连续帧中像素的强度变化具有一致性的特点，通过Harris检测算法找到强度变化剧烈和图像边缘的特征点，根据光流场与运动场的对应关系由成像平面中光流的变化估计烟雾的相对运动，计算运动矢量信息，实现多特征烟雾检测。该算法是基于烟雾灰度变化的特征点作为检测对象，大大减少了待处理的数据量，缩短了算法处理时间，综合了烟雾的局部特性和全局特性，具有较强的鲁棒性和较高的检测准确率。

关键词: Harris, 特征点, Lucas-Kanade目标跟踪, 烟雾特征, 烟雾识别