A method for pre-determining edges in 3D point clouds using contour detectors in images

Abstract. In the observation of architectural objects, the point cloud is constantly gaining popularity as a tool for quantitative analysis and visualization. Photogrammetric technologies are used to capture and create a point cloud for the object in a digital environment. A large part of the objects is created by human activity with a regular shape and structure with clearly distinguishable edges. These edges can be perceived as structural elements in the digital representation of the captured object. Their determination is a basic task in the vectorization of the resulting point clouds. The article proposes a method for isolating the part of the point cloud representing the structural elements of the object. For this purpose, based on the photogrammetrically captured images and predefined and automated processing, masks are created that isolate only the contours and edges. Thus, during the alignment of the images and the formation of the spatial model, only key points belonging to the corresponding structural elements are identified. To create the masks, the radiometric parameters of the images are modified, and a contour detector is applied to identify the projections of the structural elements in each image. The masks are saved as binary images, while the degree of the contours is refined so that enough points in the cloud are calculated to represent the structures of the object. To evaluate the results, direct measurements of some elements of the object are made. The experiments also show good results when validating with the point cloud for the whole object.