| Title |
A Mark Line-Based ORB-SLAM Framework for Improving Localization Accuracy in Construction Sites |
| Authors |
Hun-Hee Cho ; Min-Ji Jo ; Jung-Ho Yu |
| DOI |
https://dx.doi.org/10.6106/KJCEM.2026.27.4.083 |
| Keywords |
Mark-Line Based ORB-SLAM; Construction Robot Localization; Monocular low-cost visual Localization |
| Abstract |
This paper presents a mark-line-assisted ORB-SLAM framework for on-site robot localization in the texture-poor, repetitive environments typical of construction sites. The key idea is to treat the chalk-snapped layout lines that are routinely produced during construction as additional linear cues, both to strengthen the robustness of conventional visual SLAM and to anchor its trajectory in the absolute site coordinate frame. The framework consists of three stages. First, control points are extracted from the CAD layout drawing. Second, standard ORB-SLAM3 tracking is coupled with line features detected by the Line Segment Detector (LSD), so that line evidence compensates when point features are sparse. Third, the SLAM frame is registered to the site frame so that the entire trajectory can be expressed in absolute coordinates. In a controlled corridor experiment, the proposed framework reduced the trajectory root-mean-square error (RMSE) from 149.20 mm to 95.06 mm ? a 36.3 % improvement ? using only a monocular smartphone camera, and without any external beacon or dedicated infrastructure. The results suggest that construction mark lines can serve as a reliable geometric and georeferencing resource for low-cost visual localization. We also discuss the limitations of the present study and outline future work on multi-sensor fusion and full 3D, multi-floor operation. |