This is the project during my research visiting at Andread Nuechter’s group. The goal is to build efficient feature for very large point cloud matching, e.g., the city data, which is a challenge due to the large size of data and the weak capability of recent descriptor in such a dataset.
To achieve the goal, it consists of two subprojects.
A new formulating of registration problem
This work provide a non-iterative one-step solution. Its paper has been submitted to CVPR2020.