摘要
烟雾探测是预见火灾隐患的重要手段,具有快速响应和非接触等优点,但现有方法在减少计算量和减少误报上尚需改进。提出了一种基于运动块追踪的视频烟雾探测方法。采用混合高斯模型对背景进行建模,提取可疑运动区域并划分运动块,根据运动块灰度变化及与周围8邻域运动块的灰度梯度,筛选需要全程跟踪的目标运动块。在随后追踪过程中,分析各个运动块的运动方向、灰度衰减变化及半透明性特征,综合各特征值对烟雾进行判断。在测试烟雾视频中,60帧内便可以探测到烟雾块,而在干扰视频中只出现了1次误报。试验结果表明该方法减少了计算量,提高了烟雾检测的准确性,降低了误报率。
The present paper is inclined to propose an innovated smoke detection method for fire disaster forecast and rescue. In the paper,we have brought about the gray level change consistency of the motion blocks according to the attenuation of the smoke and the mixture Gaussian model to be used in the process of extracting suspicious area. The said suspicious movement areas here mentioned can be divided into the target motion blocks that are to be selected and then the observation can be done by tracking all the way in the successive frames according to the gray level change or the variance of the gray level gradient in the eight neighboring blocks. At the same time,it is necessary to trace and analyze the characteristic features of the smoke motion direction,the gray intensity attenuation change and their translucency that is likely to bring about. In so doing,it would be possible to accumulate the motion orientation and while observing assessing and calculating the high and low frequency energy intensity through the wavelet transformation situation in the tracking process. And,finally,it would be possible to determine whether the motion blocks are smoke blocks by synthesizing all the smoke eigenvalues. In so doing,it would be possible to track the image moving process by detecting the movement trend of the smoke blocks before no more than 60 frames in the test smoke video,in which the smoke frame may seem to be randomly distributed as the first frame. And,then,there would appear a false alarm in the interference videos of driving a car with the light on at night of 10 000 successive frames. The interference videos also include the leaves swaying video,basketball playing video. The experiment results of the above experiments may show that this method can help to detect the smoke blocks timely in the smoke video,which can help to facilitate the calculation process and increase the accuracy of the smoke detection,thus reducing the rate of false forecast.
出处
《安全与环境学报》
CAS
CSCD
北大核心
2016年第4期96-100,共5页
Journal of Safety and Environment
关键词
安全工程
烟雾探测
混合高斯模型
运动跟踪
灰度衰减
半透明性
safety engineering
smoke detection
mixture Gaussian model
motion tracking
grayscale attenuation
translucency
