Odhad a sledování póz netexturovaných objektů

Abstract

The goal of my diploma thesis was to implement an algorithm for detection and pose estimation of a known, textureless object in real time including a graphical output. The use of both color and depth data was also a requirement. The state-of-the-art analysis gives an overview of methods that solve this problem, from which I picked one based on the requirements and implemented an object detection pipeline based on the LINEMOD and Iterative Closest Point algorithms. Testing was mainly done on own, more complex objects than in the original work, especially in terms of robustness to occlusion and poor quality of the depth image. A test against the LINE-2D algorithm, which uses only color data, was also performed. The detections were successfully achieved at required speed and some flaws of the chosen approach were also revealed.