基于累积差分图像和高斯背景统计模型的烟尘目标检测

根据烟尘目标灰度与天空背景灰度很接近,目标内部灰度分布不均匀,且形状和面积会随着时间不停变化等特点,提出了一种由粗到精的目标检测算法。首先,通过累积差分图像序列得到烟尘目标出现的大致区域,再通过构造高斯背景统计模型得到每一个可能目标点的精确分割阈值,用该阈值在上一步操作得到的目标区域内部进行逐帧精细分割,得到烟尘目标出现的位置信息和时间信息。实验结果表明,该方法抗干扰能力强,能够可靠、快速地检测出烟尘目标。

Smoke Target Detection Based on Accumulated Differential Image Sequence and Gaussian Background Statistical Model

The smoke target's grey scale is very close to the sky background,the inner pixels' grey scales vary in a large range,and the smoke target is always change its contour.To cope with this problem,a new algorithm for smoke target detection from coarse scale to elaborate scale is proposed.First,by accumulated differential image method areas of interest can be find out,which indicates the rough area of the smoke target.Then by using Gaussian background statistical model,every potential target's segmentation threshold can be calculated.Segmentation operation can be done in the rough area of the smoke target only.Experimental results indicate that the method can efficiently and reliably detect smoke target with complex background.