Detection and Pose Determination of a Part for Bin Picking
Detekce součástky a určení její polohy pro úlohu vybírání
Authors
Supervisors
Reviewers
Editors
Other contributors
Journal Title
Journal ISSN
Volume Title
Publisher
České vysoké učení technické v Praze
Czech Technical University in Prague
Czech Technical University in Prague
Date
Files
Abstract
Tato diplomová práce se zabývá tématem vizuální úlohy vybírání, ve které se postupně vybírají součástky z bedny. Částečně strukturovná varianta této úlohy je uvažována. V této diplomové práci se navrhuje řešení této úlohy, které je záloženo na konvolučních neuronových sítích. Díky tomu, programová specifikace geometrie součástky není nutná. Návrhovaný systém odhaduje pozici a orientaci součástky a detekuje překrytí součástek. Systém byl implementován a otestován s použitím kovové součástky. Odhadovaná kvalita systému je 95 % úspěšných pokusů vybrání.
This thesis discusses the visual bin picking task, which is the task of sequential unloading a bin one part at a time using a camera as a primary source of information. The semi-structured variant of the bin picking task is considered. In this thesis, a solution for this problem that is based on learning the appearance model of a part using convolutional neural networks is proposed. Thus, no hard-coded geometry of a part is required. The models in the developed system predict the poses of the parts and detect occlusions. The proposed system has been implemented and tested with a metallic strut bracket. The experiments have shown that the achieved estimated success rate of the system is 95% of acquiring attempts.
This thesis discusses the visual bin picking task, which is the task of sequential unloading a bin one part at a time using a camera as a primary source of information. The semi-structured variant of the bin picking task is considered. In this thesis, a solution for this problem that is based on learning the appearance model of a part using convolutional neural networks is proposed. Thus, no hard-coded geometry of a part is required. The models in the developed system predict the poses of the parts and detect occlusions. The proposed system has been implemented and tested with a metallic strut bracket. The experiments have shown that the achieved estimated success rate of the system is 95% of acquiring attempts.