Vis enkel innførsel

dc.contributor.advisorSkretting, Karl
dc.contributor.advisorThorhallsson, Torfi
dc.contributor.authorYtterstad, Sigurd
dc.date.accessioned2020-11-11T09:47:59Z
dc.date.available2020-11-11T09:47:59Z
dc.date.issued2020-06-28
dc.identifier.urihttps://hdl.handle.net/11250/2687298
dc.descriptionMaster's thesis in Automation and Signal Processingen_US
dc.description.abstractThere is a significant increase in e-commerce, and groceries are entering the online platform. With this development, and with customers wanting to change their shopping habits from brick-and-mortar stores to the online platform, automation is needed to relieve the manual labor required for picking products and making order fulfillment effective. The objective of this thesis is to create a solution for picking products out of distribution containers and examine the challenges and limitations of the proposed solution. The system installation used in the approach is a robot with five degrees of freedom, three for navigating the X, Y, Z-coordinates, and two for rotating the end-effector. The proposed solution is developed with some assumptions. Two of these being that the products are not stacked in height and that the cardboard is not covering the top of the product. The approach is to capture a depth image of the scene and to apply a scale-invariant feature transform to detect and create a bounding box of the product. The region contained by the bounding box is compared to a reference image, and the color differences of the images are used for cardboard estimation. With the estimated cardboard, combined with the depth information from the camera, a collision map is created for collision detection. Two experiments are conducted. In the first experiment, the products are reset to the initial states for each pick, and a path planner based on rapidly exploring random tree is used to create the robot’s path for retrieving the product. The second experiment is based on the same approach, but the product is skewed and picked with Cartesian control. From the results, and within the assumptions and constraints of this thesis, a Cartesian control is sufficient for retrieving the products, and the cardboard estimation proves robust for a delimited range of products. However, future analyses are needed to determine the range of products the solution applies to, and it is suggested to do some research into deep neural networks, to see if it can outperform the proposed solution.en_US
dc.language.isoengen_US
dc.publisherUniversity of Stavanger, Norwayen_US
dc.relation.ispartofseriesMasteroppgave/UIS-TN-IDE/2020;
dc.rightsNavngivelse 4.0 Internasjonal*
dc.rights.urihttp://creativecommons.org/licenses/by/4.0/deed.no*
dc.subjectcomputer visionen_US
dc.subjectroboten_US
dc.subjectpath planningen_US
dc.subjectpoint cloudsen_US
dc.subjectobject pickingen_US
dc.subjectobject detectionen_US
dc.subjectinformation technologyen_US
dc.subjectautomationen_US
dc.subjectsignal processingen_US
dc.subjectsignalbehandlingen_US
dc.subjectinformasjonsteknologien_US
dc.titlePicking Products from Distribution Containers by Object Detection and Occlusion Estimationen_US
dc.typeMaster thesisen_US
dc.subject.nsiVDP::Teknologi: 500::Informasjons- og kommunikasjonsteknologi: 550en_US


Tilhørende fil(er)

Thumbnail

Denne innførselen finnes i følgende samling(er)

Vis enkel innførsel

Navngivelse 4.0 Internasjonal
Med mindre annet er angitt, så er denne innførselen lisensiert som Navngivelse 4.0 Internasjonal