为了提高现有火焰视频探测技术的有效性,对火焰候选区域在时间和空间上的特征进行了提取,引入并改进了统计地形特征的纹理描述方法.基于神经网络算法,分析了共计57个火焰图像特征,如颜色、边界粗糙度和圆形度,通过将多特征进行融合以达到准确快速地识别火焰图像的效果.笔者利用多种场景下的视频录像以及现场实时监控场景进行了系统性能测试,结果表明,本研究提出的火灾视频探测系统具有较快的图像处理速度和较高的探测率.
To improve the effectivity of existing fire image detection technology, this study extracts the spatial and time features from the candidate flame region, and further develops a pattern analysis method for the statistical land-scape feature of the flame region. A total of 57 flame features, such as flame color characteristics and boundary roughness and roundness, are analyzed by a BP neural network algorithm to fuse the multi-features and then to rec-ognize fire occurrence more accurately and quickly. In this paper, using a variety of scenarios and real-time video scene for the system performance test, results show that, the proposed fire video detection system can improve im-age processing speed and have higher accuracy.