3-D Color Depth Mapping System via 2-D LiDAR and Camera Fusion
IEEE ACCESS, cilt.14, ss.166968-166985, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 14
- Basım Tarihi: 2026
- Doi Numarası: 10.1109/access.2026.3747790
- Dergi Adı: IEEE ACCESS
- Derginin Tarandığı İndeksler: Scopus, Science Citation Index Expanded (SCI-EXPANDED), Compendex, INSPEC, Directory of Open Access Journals
- Sayfa Sayıları: ss.166968-166985
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
This study presents a low-cost 3D depth mapping system that detects and localizes objects in 3D by integrating a 2D LiDAR on a servo-driven tilt mechanism, a monocular camera, and an open-vocabulary segmentation model of the YOLO11 family. LiDAR data are converted into a coherent point cloud via a kinematically compensated sweep model, and the points are colorized and semantically labeled by a distortion-aware pinhole projection onto the segmentation mask. Parallax bleeding is suppressed by a three-stage occlusion-aware depth filter and a DBSCAN clustering step, after which each object is localized with a trimmed Axis-Aligned Bounding Box. In the 1–8 m reference box test, the mean front-face range error is 0.82 cm (maximum 2.99 cm at 3 m), while the box width and height deviate by 2.59 cm and 4.20 cm. Across six scenes containing three objects of different classes, sizes, and surface types, accuracy is unaffected by clutter or by a fifteen-fold reduction in illumination, and detection degrades only at 1% of the nominal image intensity. The occlusion filter retains 100% of the target foreground points while removing 13–18% of the points as occluded, reducing the front-face error from 16.01 to 1.72 cm under mutual occlusion. An ablation over twelve configurations identifies the publish-time