Registration of Aerial and Ground Images for 3D Reconstruction Using Tie Point Filtering Based on Reprojection Errors

To build a 3D model, drones are widely used for data acquisition due to their broad coverage and high convenience. However, it is challenging to acquire data from all areas using drones alone. Studies have been conducted to integrate ground images with drone images, but there are difficulties in registration due to the differences in view angles. This study proposes a registration of aerial and ground images for 3D reconstruction in complex environments using reprojection error-based tie points filtering. Aerial images were collected using drones, while ground images were obtained with a Terrestrial Laser Scanner (TLS). We performed Structure from Motion (SfM) using both drone and TLS images. The generated tie points were filtered based on reprojection error to obtain high-quality tie points, which were then used for registration between the drone and TLS data. We built the 3D model more richly using the proposed method and verified the accuracy according to filtering parameters and image matching algorithms.