Traces on rock outcrops are typically the primary data source for the structural characterization of the rock mass. Obtaining precise data about traces is essential because they are sometimes the only visible and measurable parameter in most outcrops. Manual recognition and detecting traces on high-resolution textured 3D models is relatively easy. One possible approach to enhance this process is to utilize machine learning to detect and measure the shape of these 3D traces. A unique pixel-wise texture mapper algorithm was developed in this paper to prepare the input data for the machine. The algorithm enables the processing of 3D mesh files by facilitating the mapping of texture pixels onto the mesh surfaces of 3D models. Subsequently, it generates a point cloud representation with the exact resolution as the original textured 3D model. Then, a virtual digital image rendering is employed to capture virtual images of selected regions. This technique helps to overcome limitations such as restricted access, lighting conditions, and shading effects caused by the rock mass's surface morphology. After AI-aided trace detection on 2D images, a three-dimensional data structuring technique was applied to the selected trace pixels. In the three-dimensional data structuring process, trace data were structured through 2D thinning, 3D re-projection, clustering, segmentation, and segment linking. Finally, linked segments were exported as 3D polylines, where each polyline in the output corresponds to a trace. An assessment of the performance of the proposed method was conducted using a 3D model of a real case study, which was used to compare AI-aided and human intelligence trace detection results. In conclusion, the proposed semi-automatic method was easy to use, fast, and accurate in detecting the dominant jointing system of the rock mass.