Integrating colored LiDAR and YOLO semantic segmentation for design feature extraction in Chinese ancient architecture
Yanyi Li, Chun Liu, Yongqi Lou, Tao Shen, Yunze Wu, Jing Guo, Ying Li, Man Zhang
Abstract
Currently, digital preservation of ancient buildings in China faces challenges including limited feature extraction methods and insufficient integration of color-semantic information. This study proposes a method integrating 3D LiDAR and panoramic images to systematically recognize and segment design elements in Chinese ancient architecture. The framework comprises three modules: (1) Scene construction using multi-sensor fused LiDAR point clouds for accurate 3D reconstruction; (2) Color feature extraction with our improved YOLO v11 network (YOLO-Center), incorporating a Context Anchor Attention mechanism to detect 14 types of design features in complex backgrounds, while assigning color masks for subsequent analysis; (3) Color fusion through virtual sphere mapping that precisely integrates image colors with point clouds. The resulting fused model preserves spatial structures while visualizing distinct design features. This approach enhances accuracy and efficiency in feature extraction, advancing semantic understanding and digital preservation for architectural restoration.