双目遥感凝视系统中运动小目标的快速自动检测

High-speed auto-detection of moving small targets in remote sensing staring binocular imaging system

从双目遥感凝视系统的视场重叠区进入系统的信息量大于通过非重叠区的信息量,根据这一特征建立了一种新的运动小目标双目并行快速实时自动检测算法。首先采用差分向量无穷范数算法对原始图像序列做预处理,去掉大量低频噪音和背景,然后采用光流场法对运动小目标进行分割,最后用所提出的空间时间并行快速判定算法对分割的可疑运动小目标进行判定。实验结果表明:由于识别判定算法的空间时间是并行处理的,所以识别判定的平均速度比单目视觉系统提高了50%;在图像信噪比不小于5dB的情况下,准确判定识别的概率为97%。

According to the feature of remote sensing staring binocular imaging system, the information quantity passed through the overlapped fidd of view is larger than that passed through the non-overlapped fidd of view, a new paralld high-speed automatic detecting algorithm of moving small targets is established. The algorithm firstly used the Infinite Norms of the difference vectors of the processed images as a preprocessing algorithm to get rid of most of low frequency noise points and background, then used Optic Flow to segment the moving small targets, and finally used the novel space-time parallel high-speed determining method proposed to determine whether the doubtful moving small target was a true target or not. The experiments results prove that the average determining speed of the space-time parallel processing determining method is improved 50 % than that of the monocular imaging system, in the SNR of processed images is no less than 5 dB, the correct determining probability is 97 %.