结合颜色和外型特征的视频烟雾识别方法  

Video smoke detection based on color and shape features

作　　者：王森[1] 葛思擘[1] 邹建华[1] 马玉洁[1] 

机构地区：[1]西安交通大学电子与信息工程学院,西安710000

出　　处：《计算机应用》2016年第A02期168-170,213,共4页journal of Computer Applications

摘　　要：针对烟雾颜色特征判别时,仅仅根据具有相关性的RGB三通道值相近这一准则造成的判别结果不准的问题,提出了一种基于RGB颜色空间三通道和Lab颜色空间a通道的四维颜色空间的烟雾颜色特征判别方法,结合烟雾的外型特征进行视频烟雾的识别。首先,采用Vi Be算法检测出视频图像中的前景目标;然后,经过形态学开闭处理和连通区域分析后,训练烟雾像素数据并计算RGBa四个通道的颜色概率初步排除伪烟块;最后,通过烟雾的轮廓复杂度和边缘密度等外型特征进一步排除伪烟块得到最终的识别结果。理论分析和实验结果证明这一识别方法兼具实时性和准确性。There exists a problem of accuracy simply by the rules that the values of RGB channels are close to each other when deciding a pixel belongs to smoke or not since these channels are related. Therefore, RGBa, an four-dimensional color space model which combined RGB channels and the a-channel of Lab color space was proposed to determine color feature. Shape features were also applied to improve the accuracy of recognition. Firstly, the ViBe algorithm was applied to detect the foreground of input video. Then, the morphological operation of opening and closing was performed, data of smoke pixels were trained, color probability of each channel was calculated to remove non-smoke area. Finally, shape features of contour complexity and edge density were taken into account. Theoretical analysis and experimental results show the proposed approach can be both immediate and precise.

关 键 词：烟雾识别 前景提取 颜色概率 轮廓复杂度 边缘密度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]