基于YUV颜色空间和多特征融合的视频烟雾检测

Video Smoke Detection Based on YUV Color Space and Multiple Feature Fusion

针对基于视频的烟雾检测方法在复杂环境中存在高误报、高漏报问题,提出了一种基于YUV颜色空间和多特征融合的视频烟雾检测算法。首先通过高斯混合模型检测运动区域,然后应用YUV颜色空间烟雾像素过滤方法确定疑似烟雾区域,进一步将疑似区域块分割后使用离散小波变换和均匀局部二值模式提取烟雾纹理特征,最后多特征融合并通过Real AdaBoost分类器进行火灾识别分类。实验结果表明,高斯混合模型能够有效检测运动区域,运动区域YUV颜色空间过滤可以滤除大量非烟雾像素区域,选取的烟雾特征及分类方法具有更高的识别精度。多个场景视频测试表明所提出的视频烟雾检测算法相比同类算法具有识别准确度更高、误报率更低,从而更符合不同的应用场景。

The application of video-based smoke detection methods still has the difficulty with high false positives,high false negatives in complex environments. To solve the problems,this paper presents a video smoke detection method based on YUV color space and multiple feature fusion algorithms. Firstly,the motion region is detected by the gaussian mixture pattern,and the suspected smoke region is determined by filter pixel in the YUV color space.Then,the suspected smoke region was divided into blocks and discrete wavelets,uniform local binary are used to analyze the texture of the block in the frequency domain and the spatial domain. Finally,the wavelet features and localized local binary features are linearly fused and identified using the Real AdaBoost classifier. The experimental results show that the suspicious smoke area was determined only contains smoke pixels,and the smog feature and feature classifier selected have higher recognition accuracy than the state-of-the-art methods. Several scene video were tested and showed that the smoke video was quickly and accurately identified with much lower false positive rate by the proposed video smoke detection method.