Processing speed improvement on SFM using panoramic images based on camera direction constraints

Point clouds are generated with structure from motion (SfM) and laser scanning for 3D modeling of objects and ground surfaces. Although SfM is a lower cost approach than laser scanning, SfM requires plenty of time for image acquisition from various positions and angles. When many images are used for dense point cloud generation, the image processing cost increase nonlinearly. Thus, SfM holds two technical issues such as image acquisition cost and image processing cost. Therefore, we firstly focused on the use of low-price omni-directional camera to improve image acquisition cost. We secondary focused on the efficiency improvement in image matching procedures in SfM processing based on camera direction constraints. We select stereo pairs to reduce processing time based on bundle constraints with image matching lists prepared in advance. We conducted experiments using panoramic images to evaluate our methodology. Through the experiments, we confirmed that our methodology can improve the efficiency of image acquisition cost and image matching processing cost.