当前位置: X-MOL 学术Autom. Constr. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Mosaicking of mountain tunnel images guided by laser rangefinder
Automation in Construction ( IF 9.6 ) Pub Date : 2021-04-28 , DOI: 10.1016/j.autcon.2021.103708
Meng Du , Jia Fan , Yuchun Huang , Min Cao

Routine tunnel inspection is important for timely maintaining tunnels and ensuring traffic safety. Owing to the insufficient field of view of one camera, an array of cameras has been employed to acquire multiple tunnel images simultaneously, which are then stitched together to obtain an overall tunnel lining image (TLI) of the complete defects and facilities on the lining surface. However, due to the features inadequacy on the poorly-textured surface, a traditional feature-based mosaicking of multiple images does not work for most tunnel images. Based on the guidance of a laser rangefinder (LRF), this study proposes stitching tunnel images from a geometric perspective involving three steps: extrinsic calibration, coarse mosaicking, and fine mosaicking. Using a checkerboard, the extrinsic parameters of the cameras relative to the LRF are calibrated. The LRF range profile is used to construct a 3D tunnel model, through which a lookup table (LUT) is generated to stitch the tunnel images geometrically, resulting in a coarse TLI. Finally, to reduce misalignments from inaccurate calibration, optional graph cuts are made to optimise the coarse mosaicking if evident features exist in the overlapped areas between images. The proposed technique is tested on actual images collected from the Dingxi tunnel and the Huang Longshan tunnel in China. The experimental results show that the proposed algorithm can obtain the TLI with little geometrical distortion and ensure structural consistency in the features, even if auxiliary facilities are densely distributed on the inner surface of tunnels. The time required for stitching the tunnel images can be improved by 82×, achieving 0.52 s per TLI, if graph cuts are not required.

更新日期:2021-04-29
down
wechat
bug