Rozpoznání objektů a jejich polohy s využitím 3D modelů objektů

Abstract

Thesis addresses one of the typical tasks in computer vision, called 3D pose estimation. Object recognition is the fundamental task for different areas, such as robotics or augumented reality. The goal of this thesis was to develop a software for 3D pose estimation in RGB-D image based on object's model using various State of the Art local descriptors algorithms. Later, these algorithms are compared and evaluated. Moreover, a few enhancements are proposed to make object recognition process more effective. These include dimension reduction using principal component analysis (PCA), adding color to the Fast Point Feature Histogram (FPFH) descriptor, modifying FPFH descriptor to estimate low dimensional descriptors and comparing algorithms for nearest neighbor search.