Tunnel image stitching based on geometry and features

Abstract In the fast tunnel detection of multi-camera, a disease distributed in multiple images is prone to be misidentified as multiple diseases, which affects the evaluation of the status of the tunnel. This paper proposes a high-precision stitching method driven by data and scene based on multi-camera sequence images. Firstly, geometric rough calculation is performed to generate the theoretical stitching mode using the geometric positional relationship between the cameras in the scene and the image relationship is renewed after theoretical stitching. Secondly, feature points are extracted and matched for adjacent overlapping images by SURF algorithm. Therefore, pixel-level data registration is performed to achieve the image stitching. Finally, an integrated stitching mode is proposed based on the theoretical stitching and pixel-level data registration, which utilizes the stitched sequence images with high physical resolution to extract cross-section information. Practical results show that the method can achieve image stitching of the tunnels with high accuracy and good reliability.