Rigorous Projection for Image Stitching: a 3D-Informed Approach for Accurate Panoramic Photogrammetry

Abstract. The paper presents a 3D-informed method for generating stitched panoramic images from multi-camera rigs through rigorous reprojection onto a scene model. Unlike conventional and parallax-tolerant stitching, the proposed approach explicitly accounts for camera calibration, relative orientation, and scene geometry, with the aim of reducing parallax effects while preserving metric consistency. The method is tested on confined environments, where non-coincident projection centres make stitching especially critical, and is evaluated with different rig configurations, including systems with both small and large sensor baselines. Two experiments are performed: a stitching-accuracy test against synthetic reference panoramas, and a Structure from Motion (SfM) test comparing rigorous panoramas with raw fisheye processing. Results show that the proposed approach yields geometrically consistent panoramas and substantial gains in processing efficiency, although raw fisheye images still provide the best overall metric performance in the most demanding reconstruction scenarios.