基于混合高斯模型和帧差法的吸烟检测算法

摘要

提出一种快速有效的基于混合高斯模型和帧差法的视频吸烟检测技术。利用混合高斯模型获取准确的背景，通过帧差分法对连续两帧的灰度图像序列进行绝对差运算，与背景相减获得运动目标区域；根据改进的转换模型将稀薄烟雾图像从RGB颜色空间转换到HSV颜色空间，对其颜色特征提取并进行分析，进一步将稀薄烟雾区域从运动目标区域中分离出来。实验结果表明，该技术不易受到周围环境的限制，灵敏度高、适用范围广。%An effective cigarette-smoke detection technology based on Gaussian mixture model and frame difference was pro-posed.The accurate background was obtained using Gaussian mixture model,and two adjacent frames of grayscale image se-quence were processed using absolute difference done through frame difference,then accurate background was subtracted to ob-tain moving obj ect region.The thin smoke images were changed from RGB color space to HSV color space according to an im-proved the transformation model and their color characteristics were extracted and analyzed,and the thin smoke region was sepa-rated from moving object region further.Experiment results show that the technology is not easily limited to environments,and it also has high sensitivity and wide usability.