Improving Mountainous DSM Accuracy Through an Innovative Opposite-Side Radargrammetry Algorithm

Abstract

Radargrammetry is a critical technique for generating a high-resolution digital surface model (DSM). In radargrammetry, a large intersection angle between stereo images leads to higher elevation accuracy. Traditional radargrammetry often utilizes same-side stereo SAR images with a small intersection angle. Opposite-side radargrammetry can achieve higher accuracy DSM with its large intersection angle. However, dense stereo matching of opposite-side images is challenging due to the different orbit directions of the satellites, especially in mountainous areas. To address this issue, we propose an innovative indirect SAR image matching algorithm for generating opposite-side radargrammetric mountainous DSM. First, a SAR image simulation method is proposed to connect the opposite-side SAR images using slope and orbit information. Second, a triangle affine matching algorithm is developed to match the simulated SAR and real SAR images based on feature points. Then, the opposite-side SAR images can be matched according to the proposed algorithm. Finally, the stereo positioning method is introduced to obtain the geographic coordinates point cloud and the final DSM. The proposed method is validated using a spaceborne GaoFen-3 dataset over the mountainous area in Omaha, Nebraska, USA. The generated DSM is compared against open-source light detection and ranging data from the U.S. Geological Survey. The results demonstrate that the proposed method achieves a root mean square error of 6.41 m, representing a 24.2% and 20.1% improvement compared to the same-side radargrammetry method and the existing opposite-side radargrammetry method, respectively.