In the field of cultural heritage dissemination, the availability of an accurate 3D model is mandatory to better describe an historically relevant building and provide the user with as much information as possible. In our paper we propose a workflow to improve the accuracy of the sparse cloud (Tie Points) by means of automated mask generation on photograms acquired by UAV (Unmanned Aerial Vehicle). In the first step, the statistical distribution of reprojection errors for each image, before applying the masks, is analyzed. In the second step, masks are automatically generated for each image to exclude, in the following alignment process, parts of the image characterized by reprojection error values outside the chosen confidence interval. The BIC criterion was used to identify the probability distribution that best fits the data. The results are promising and highlight the need to implement specific processes for preparing the input data to improve the accuracy of the output 3D model. Such results are not easily achievable with the processes implemented by default in the most widely used commercial photogrammetric software.
Abstract. The paper presents an innovative approach that can assist survey methods by applying AI algorithms to improve the accuracy of point clouds generated from UAV images. Firstly, the work individually analyses several photogrammetric accuracy parameters, including reprojection error, angle of intersection between homologous points, number of cameras for single Tie Point calculation, verifying that a single parameter is not sufficient to filter noise from a photogrammetric point cloud. Therefore, some of the calculated parameters were analysed with the Self-Organizing Map (SOM) and a K-means, to check the impact of the precision parameters for reducing the noise associated with the definition of the 3D model. In the case study, in both machine learning clustering algorithms used, it was observed that the parameter that most influences noise in photogrammetric point clouds is the angle of intersection.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.