Structure-from-Motion Retrieval from VideoSAR Sequences: Toward a Unified Framework

Abstract

Video synthetic aperture radar (VideoSAR) has attracted significant attention in the computer vision-based information retrieval field. In this paper, a unified three-dimensional reconstruction framework is proposed for both the straight and circular VideoSAR modes, and is suitable for height retrieval with/without geometric prior knowledge. First, two VideoSAR imaging modes are presented to explain the formation of dynamic observation. Then, structure-from-motion retrieval method with/without geometric prior knowledge is proposed by exploiting shadow information. Finally, experimental results based on VideoSAR fragments measured by the airborne and MiniSAR systems demonstrate the effectiveness of the unified retrieval method with a higher accuracy compared with that of a single SAR imagery.