- Marc Morenza
- Victor Casamayor
This project was developed as part of the master thesis "Robot 3D mapping: Tunning and Assessment" of UPF's master in Intelligent Interactive Systems in 2017.
THE REPOSITORY IS NO LONGER MAINTAINED, BE AWARE OF THIS IF YOU TRY TO USE IT
This package uses:
This software will compare two input pointclouds, providing with some metrics about their relationship. Number of points, descriptors, noise and number of elements based on color differences are the current metrics. The program finds matches of clusters between each point cloud and then compares those matches, it also shows each of the cluster of each match on screen. Additionally, there is a "results.txt" file in "./build" in which appears the information of the segments along with the results of the comparison of each match. At the end of the file there are some scores, telling how much different are the matched segments regarding the previously mentioned metrics. Additionally, two similarity ratios between both point clouds are computed, based on the results of previous scores. The first similarity states, in general, how much similar are the clusters of the first point cloud to the clusters of the second one regarding the scores. A number higher than one will say that the second point cloud has more information, whereas a number less than one will say that the second has more. The closer this number is to one, the more similar the clusters matched will be. The second ratio gives a measure of how much similar the general point clouds are between them, basing on how much similar are the matched clusters and how much portion of the first point cloud is similar to the second point cloud.
In order to build this program, it would be necessary to have already installed the PCL library along with their third party libraries if necessary. The version used here is 1.7, be aware that previous versions of pcl library might not work. After the installation of the library, you need to modify the CMakeLists.txt with your own paths. Once done, run from "./build" the command "cmake ..", which will tell you if there is a problem with CMakeLists.txt. If there is no error, you can run from "./build" the command "make" and it will compile the code and create the executable "comparator".
PLEASE BE AWARE THAT THIS IS EARLY DEVELOPMENT.
In order to use it, you will run "./comparator /PathToPointCloud1 /PathToPointCloud2" from "./build". By default, the visualization of the clusters and the matches is off, you can run with "-v": "./comparator -v /PathToPointCloud1 /PathToPointCloud2" from "./build" if you want it active. Similarly, by default noise analysis is disabled as it takes quite more time. If you want to run with it then run with "-n": "./comparator -n /PathToPointCloud1 /PathToPointCloud2" from "./build". You might run as well both "-n -v". Help is also available with "-h". Other options are -e to select euclidean cluster segmentation as main segmentation algorithm and -i to perform ICP before the point clouds comparison to check if they are comparable enough. If you activated the visualizer, IN ORDER TO EXIT THE VISUALIZER PRESS "Q". The programm will output how it is doing on the command line. If it has found a match between both point clouds, they will appear if you activated visualization. You need to press "Q" again for the programm to continue for each viewer. All results will appear in the file "results.txt".
-
Here it is a link with some point clouds used for the testing. There are several versions of each point cloud (noisy, downsampled etc) plus some other general pcls: google drive folder with pcls.
-
Here it is a link with the results files of the program with several tests with the point clouds of the previous folder: Google drive folder with result files