Quality features for the integration of terrestrial and UAV images

The paper presents an innovative approach for improving the orientation results when terrestrial and UAV images are jointly processed. With the existing approaches, the processing of images coming from different platforms and sensors leads often to noisy and inaccurate 3D reconstructions, due to the different nature and properties of the acquired images. In this work, a photogrammetric pipeline is proposed to filter and remove bad computed tie points, according to some quality feature indicators. A completely automatic procedure has been developed to filter the sparse point cloud, in order to improve the orientation results before computing the dense point cloud. We report some tests and results on a dataset of about 140 images (Modena cathedral, Italy). The effectiveness of the filtering procedure was verified using some internal quality indicators, external checks (ground truth data) and qualitative visual analyses.