Abstract: A trace is a spatial curve formed by the intersection between rock joints and the rock-mass free surface. Its geometry directly reflects the rock-mass structural characteristics. Therefore, the rapid and accurate extraction of trace information is of both theoretical and practical engineering significance. Current trace-extraction methods for spatial point clouds rely primarily on curvature, and rarely incorporate point-cloud color information. Moreover, current trace-extraction methods for spatial point clouds rely primarily on curvature, and rarely incorporate point-cloud color information. To address these limitations, we propose a new trace line extraction method based on dimensionality reduction projection (NTDR). NTDR conformally projects 3D point clouds onto a 2D plane and performs efficient edge detection using the color information in the projected data. It then clusters and connects candidate points by integrating 3D geometric features (e.g., curvature and point-to-point distance), enabling automatic extraction of joint traces. Experimental results show that, for large-scale point clouds, NTDR reduces processing time by 91.07% relative to manual extraction, substantially improving efficiency. The traces extracted by NTDR achieve a 90.42% overlap with manually extracted traces and preserve more local details, indicating improved accuracy and overall performance. NTDR maintains an identification accuracy above 80% with 20% noisy points, indicating robustness to noise. Compared with similar automated methods, NTDR yields better extraction results and more closely matches the observed trace distribution in the rock mass. NTDR can improve the efficiency of geohazard prediction and provide data to support tunnel-support design and engineering safety assessment.

Key words: 3D point clouds, feature detection, trace extraction, dimensionality reduction projection