一种VideoSAR动目标阴影检测方法

Approach to Moving Targets Shadow Detection for VideoSAR

在高帧率的视频合成孔径雷达(VideoSAR)成像模式获得的图像序列中,多普勒频移使运动目标在实际位置留下阴影,且相邻帧图像具有很强相关性。该文针对上述现象提出一种VideoSAR图像中动目标阴影检测的方法。首先,对每帧图像通过结合尺度不变特征变换(SIFT)和随机抽样一致性(RANSAC)算法实现配准并进行背景补偿,再采用CattePM模型抑制相干斑噪声。然后通过Tsallis灰度熵的最大化阈值分割方法自动分离目标和背景,获得二值图像。最后,对相邻多帧图像背景建模并差分,再结合三帧间差分法提取动目标阴影,结果标记至原帧图像相应位置。基于美国Sandia实验室公布的VideoSAR成像片段,实现了多个移动车辆的检测,验证了所提算法的有效性。

In the image sequence obtained by the high frame rate Video Synthetic Aperture Radar （VideoSAR） mode, the Doppler shift results in some shadows of the moving targets in their actual position, and a strong correlation exists between adjacent frames. Based on the above rationale, this paper proposes an approach to detecting moving targets＇ shadow in VideoSAR imagery. First, the Scale-Invariant Feature Transform （SIFT） with RANdom SAmple Consensus （RANSAC） registration algorithm is used to compensate for the change of background of each frame, and the CattePM model is employed to suppress the speckle noise effectively. Then, in order to separate the targets and the background and generate binary images automatically, a threshold segmentation algorithm, called maximizing the Tsallis entropy, is applied. Finally, shadow detection is accomplished by the background difference with three frame difference method, and the detection results are marked on the corresponding position in the original frame. Experimental results utilizing the VideoSAR imaging fragment published by Sandia National Laboratories show that multiple moving vehicles are detected effectively, hence the validity of the approach is demonstrated.