A New Multi-Criteria Tie Point Filtering Approach to Increase the Accuracy of UAV Photogrammetry Models

Abstract

Extracting accurate tie points plays an essential role in the accuracy of image orientation in Unmanned Aerial Vehicle (UAV) photogrammetry. In this study, a Multi-Criteria Decision Making (MCDM) automatic filtering method is presented. Based on the quality features of a photogrammetric model, the proposed method works at the level of sparse point cloud to remove low-quality tie points for refining the orientation results. In the proposed algorithm, different factors that affect the quality of tie points are identified. The quality measures are then aggregated by applying MCDM methods and a competency score for each 3D tie point. These scores are employed in an automatic filtering approach that selects a subset of high-quality points which are then used to repeat the bundle adjustment. To evaluate the proposed algorithm, various internal and external studies were conducted on different datasets. The findings suggest that our method is both effective and reliable. In addition, in comparison to the existing filtering techniques, the proposed strategy increases the accuracy of bundle adjustment and dense point cloud generation by about 40% and 70%, respectively.