In the master's thesis, we explored the functionality and performance of functions for the localization of products in bulk as provided by the open-source PCL library. The library offers a larger number of functions and consequently multiple possible solutions, that is why we carried out experiments to evaluate their accuracy and operating speed. We tested different software solutions by searching for specific metal objects placed on a conveyor belt. During testing we changed the layout and degree of obscurity of the searched for objects. The results of the experiments show that by choosing the correct combination of the library's functions and their parameters, we significantly influence the computational speed and accuracy of the localization process. We detirmened that the best combination of functions was using Harris3D for determining keypoints, SHOT for calculating descriptors and GC for correspondence grouping.
|
