结合Kmeans++聚类和颜色几何特征的火焰检测方法

Fire detection method using Kmeans++clustering and features of mixed color and geometry

随着社会经济的不断发展,大空间建筑逐渐步入人们的生活中,对大空间建筑的消防技术要求逐渐提高,火焰识别技术已成为近年来研究的热点。为实现单帧图像的火焰检测,本文首先提出了一种基于RGB和HSI颜色模型的混合判据,它既保留了RGB模型中的直观判据,又加入了HSI模型中对于饱和度判据,效果优于两者单独使用或单纯结合的情况;同时利用基于加权欧式距离的方法对图像进行特殊灰度化处理,通过Kmeans++颜色聚类,完成火焰图像的分割,获得最终感兴趣区域;提取该区域几何轮廓并利用不规则度和形态比例等几何判据,对待检测图像进行最终的识别。为评估所提出检测方法的性能,选取典型火焰图像和非火焰图像,在Visual Studio 2013环境下进行对比实验,通过对运行时间、提取偏差率和识别误报率等结果的分析,证明了所提方法的有效性和可实现性。本文所提出的方法具有良好的检测效果,能够保证火焰提取和识别的精度,同时兼顾实时性的要求,可以应用在实际的大空间消防项目中。

With the continuous development of social economy, large space buildings have gradually entered people' s lives, and the requirements of fire protection technology have gradually increased. Therefore, fire detection has become a hot spot of research. In order to realize the fire detection of single image, firstly, the mixed color criterion with RGB and HSI criterion is used to form the initial region of interest. Based on the Kmeans++ clustering algorithm, the fire image is clustered after the special grayscale processing. The clustering result is combined with the initial region of interest to obtain the final region.The contour of the region is extracted, and the image is identified by geometric criteria such as irregularity and shape ratio. Finally, to evaluate the performance of the proposed method, experimental research under Visual Studio 2013 environment is performed. The experimental results show that the proposed method have low extraction deviation rate and false positive rate. It has good effectiveness and can be applied to fire protection technology of large space.