Abstract

The quality of models produced by photogrammetry applications is affected by various factors. Two important factors are pose and image quality. Blur artifacts are omnipresent in most datasets to some extent, especially when acquired through video scans. These artifacts can worsen the reconstruction result. Analogously, falsely calculated camera poses can also affect the resulting models negatively. In this guided research, we examine how camera poses calculated from COLMAP can be evaluated using ground-truth estimations provided by Augmented Reality (AR) applications. Furthermore, we present an algorithm to detect blurred images in photogrammetry image-sets by comparing images to other photographs in its vicinity. To do so, we introduce different approaches, which were explored over the course of this guided research, in an effort to encapsulate the concept of image neighborhood. In the end, we briefly evaluate the algorithm and reconstructions made with the filtered datasets. The results indicate a slight improvement of the model textures, though further refinement of the filtering algorithm is required to prevent the deletion of unique viewpoints

Results/Implementation/Project Description

Conclusion

[ PDF (optional) ] 

[ Slides Kickoff/Final (optional)]