Prepoznava in lokalizacija izdelkov na osnovi 3D slik

Matjan, Rok (avtor), Bračun, Drago (mentor)

Na področju avtomatizacije in uvajanja robotskih sistemov v proizvodnjo je pogost problem pobiranje naključno razporejenih izdelkov s traku ali iz zabojnikov. V tem delu se spoznamo s tehnologijami zajema 3D slik in njihovo obdelavo s ciljem prepoznave in lokalizacije izdelkov. Razvito je preizkuševališče, ki vključuje tekoč trak, na katerem so naključno razporejeni izdelki, 3D merilnik, ki zajame 3D sliko v obliki barvnega oblaka točk, ter industrijski računalnik, ki krmili celoten sistem in obdeluje oblake točk. Za obdelavo 3D slik je uporabljena metoda, kjer najprej oblak točk filtriramo, segmentiramo in na njem izračunamo opisnike. Ti se nato primerjajo z opisniki iskanega objekta in v kolikor sovpadajo, potrdimo obstoj iskanega objekta in določimo njegovo lego. Razvita metoda je verificirana na primeru lokalizacije ulitkov na tekočem traku.

• Jezik: Slovenski jezik
• Ključne besede: lokalizacija izdelkov, 3D slika, knjižnica PCL, segmentacija, registracija, prepoznava
• Vrsta gradiva: Diplomsko delo/naloga
• Tipologija: 2.11 - Diplomsko delo
• Organizacija: FS - Fakulteta za strojništvo
• Kraj izida: Ljubljana
• Založnik: [R. Matjan]
• Leto izida: 2023
• Št. strani: XX, 48 str.
• PID: 20.500.12556/RUL-145778
• UDK: 004.925:007.52(043.2)
• COBISS.SI-ID: 152083203
• Datum objave v RUL: 13.05.2023

Sekundarni jezik

• Jezik: Angleški jezik
• Naslov: Recognition and localization of products based on 3D images

Izvleček

In the field of automation and the introduction of robotic systems in production, a common problem is to pick randomly distributed products from a conveyor belt or from bins. In this work, we address 3D image capture and processing technologies with the goal of product identification and localization. A testbed is developed that includes a conveyor belt on which products are randomly arranged, a 3D measurement device that captures the 3D image in the form of a coloured point cloud, and an industrial computer that controls the entire system and processes the point clouds. The method for processing the 3D images is to first filter the point cloud, segment it, and calculate descriptors. These are then compared with the descriptors of the search object and if they match, the existence of the search object is confirmed and its position is determined. The developed method is verified using the localization of castings on a conveyor belt as an example.

• Ključne besede: object localization, 3D image, PCL library, segmentation, registration, recognition