Stationary Vehicle Detection Using Complementary Information in Optical-SAR Images
Yanjie Zhang, Jinshan Ding · IEEE Transactions on Geoscience and Remote Sensing · 2024
Target detection in remote sensing images has received much research interest in the past years. Many saliency detection methods and constant false alarm rate (CFAR) detection methods have been proposed, which work effectively in simple scenes and high-resolution images, and unfortunately, they deteriorate in cases of complex scenes or low imaging quality. Multisource remote sensing images become much more accessible these years, and they provide redundant and complementary information about the area of interest. This article presents a detection method for stationary or slow-moving vehicle targets in complex scenes using optical and synthetic aperture radar (SAR) images. We utilize the contrast pyramid (CP) algorithm to fuse optical-SAR images and employ distinct methods to detect the saliency probabilities of optical, SAR, and fused images, respectively. Then, we calculate potential target areas based on these saliency probability maps and employ a region-growing method within these areas to extract the contours of possible targets. Finally, a CFAR detection method is used to obtain accurate vehicle targets. Experimental results demonstrate that the proposed method significantly improves the detection performance of stationary and slow-moving targets in complex scenes.