基于3D Max-median和3D Max-mean滤波的红外小目标检测

Detection Algorithm of Small IR Target Based on 3D Max-median and 3D Max-mean Filter

针对复杂背景下红外小目标的实时检测需要,提出了一种基于时空域条件下改进性的最大中值滤波和最大均值滤波的检测算法。首先,输入红外序列图像,再用沿时间轴和空域组成时空三维组合帧,根据连续3帧图像形成3×3×3的结构,并根据利用最大中值或者最大均值滤波来确定中间帧中心位置的像素值;然后,再用阈值分割出目标;最后,归一化逐帧叠加确定目标运动轨迹检测出目标。实验结果表明,在不同背景条件下,该检测算法能有效地检测出运动小目标,并且能准确地得到目标的位置,适合于实时的目标检测。

For real-time infrared small target detection under complex background, an improved spatial-temporal max-median and max-mean filters method is presented in this paper. Firstly, a 3 by 3 by 3 spatial-temporal structure is constructed by using consecutive three frames from sequential ima- ges and the center pixel value is determined by max-mean or max-median filter result. Secondly, the potential targets are obtained by segmentation. Fi- nally, the final targets are determined by target motion trajectories which are achieved by accumulating normalized frames. Experimental results show that the proposed method can effectively detect motion small targets and determine the target locations. In addition, this method is suitable for real-time de- tection tasks.