Feature Line Extraction from 3D Model of Oblique Photogrammetry Based on Multi-Objective Weighted Shortest Path

Abstract

Oblique photogrammetry 3D models suffer from coarse structure, high noise and regularity deficiency in corner or ridge regions, which make it difficult to extract feature lines quickly and accurately from these regions. To deal with this problem, a method of feature line extraction from 3D model of photogrammetry based on multi-objective weighted shortest path is proposed. First, the model is pre-processed to build a complete and continuous topological structure, and organized as a weighted directed graph. Then, considering the distance, direction and the change trend of the triangulation, the weights are calculated; the Dijkstra algorithm is constrained to obtain the shortest path to get the feature lines. Finally, using the feature line extraction results, a method is proposed to repair the regions without distinct features. Results show that compared with the interactive method, the proposed method is efficient and only need select two feature points to specify the target. At the same time, the extraction results do not rely on artificial experience and are highly objective. Compared with the automatic extraction method based on edges and faces, this method is less affected by noise, and can extract the specified feature line under simple interaction.