Rozpoznávání objektů na základě jejich 3D modelů

Abstract

The thesis deals with 3D object recognition from point cloud. Point clouds can be obtained in two ways, either by using depth sensors or by generating them synthetically. Practical applications of this topic are in autonomous driving, robotics, augmented reality and other emerging areas. The aim of this work is to develop an experimental program that compares different keypoint and descriptor methods that are implemented in the open-source PCL library. Subsequently, the chosen methods will be tested on synthetic scenes. Finally, a recapitulation of the findings and results obtained will be presented.