中文摘要：

为更好地提取火焰的动态特征,提高火焰视频识别的准确率,提出一种利用旅行者行为图论的火灾火焰视频动态纹理分析与识别方法。把图像像素点看作景点,建立火焰视频的旅行者行为图,定义火焰视频纹理分布函数;在时空域推导火焰动态纹理特征向量公式,详细阐述参数的选择方案;用支持向量机进行火焰视频识别。该算法性能不依赖于疑似火焰区域的检测,从视频片段大粒度角度分析火焰的动态特征。实验结果表明,该算法降低了算法耗时,火灾视频识别率较高。

英文摘要：

To effectively extract dynamic features and enhance the recognition performance of flame video,a dynamic texture analysis algorithm of flame video based on tourist walk graph was presented.Each pixel of one flame image was regarded as one spot to establish tourist walk graph of flame video.Based on this graph,texture distribution function was defined.The flame feature vector formulas in spatio-temporal domain were derived,and the parameter selection scheme was elaborated.SVM was used to recognize flame video clip.The performance of the proposed algorithm was not dependent on detection accuracy of suspected flame region,and the dynamic characteristics of flame video were analyzed in term of large granularity of video clip.The experimental results show that the proposed algorithm has higher accuracy on flame video recognition with comparable computational time.