Using Augmented Measurements to Improve Convergence of ICP
This web page contains all the data (or equivalently a reference) generated by NICP, GICP and ICP algorithms over the fr1/360, fr1/desk, fr1/desk2 and fr1/room RGB-D datasets.
Each .png plot image is named as follow:
where the transl and rot identify respectively plots showing the translational error and the rotational error for that particular alignment. The error at each iteration of a plot represents how much far is the computed transform from the real one.
Similarly as for plot images each .txt file is named as follow:
This .txt files contain the translational and rotational error, for all of the algorithm compared, at each alignment iteration. In particular from left to right each column represents:
NICP_translational_error NICP_rotational_error GICP_translational_error GICP_rotational_error ICP_translational_error ICP_rotational_error
There are particular .txt files that are named as follow:
and they contains all the transformations found in each alignment from each compared algorithm. In particular from left to right each row contains the number of alignment and for ground truth, NICP, GICP and ICP the translation (as a 3 float elements vector tx, ty, tz) and the rotation (as a 4 float elements quaternion qw, qx, qy, qz).
It is possible to run locally the simulated experiments just following these steps:
download and unpack the file posit_solver.tar.gz
chmod +x mkexperiments.sh
Since from the google drive folder view it is not possible to visualize more than 500 files, you need to add the folder in your drive (upper right button) and open it in your google drive (again upper right button). If you don’t want to have a google drive account you can still download the data, plots and code by downloading the files “data+plots.zip” and “posit_solver.zip” caontained in the main directory: “Using Augmented Measurements to Improve the Convergence of ICP”.