Visual Object Pose Estimation

Contact Persons

Pengyuan Wang (pengyuan.wang@tum.de), Fabian Manhardt (fabian.manhardt@tum.de), Benjamin Busam (b.busam@tum.de)

Project Coordinator: Benjamin Busam

Abstract

Our team focuses on the task of 6D object pose estimation. In particular 6D pose estimation attempts to infer the 3D location and orientation of the object of interest. We actively collaborate with several partners from industry, including Toyota Motor Europe, Siemens, etc. Within these cooperations, we leverage our research expertise to develop solutions for indoor robotic applications for different sectors such as healthcare. Thereby, our team constantly pushes the envelope of what is thought to be possible. Exemplary, we introduced the first deep-learning driven approach for purely monocular 6D pose estimation. Certainly, we also explore several directions and applications to improve the applicability of pose estimation. For instance, we work on improving the generation of synthetic training data to enable large-scale training on multiple objects. We also investigate class-based object detection, involving the estimation of the 3D shape and 6D pose of unseen objects, rather than limiting the applications to a limited setup of handling only a handful objects.

Keywords: 3D Computer Vision, 6D Pose Estimation, Robotic Manipulation

Research Partners

Toyota Motor Europe NV/SA, Belgium, Siemens AG, Munich

Project Members

Slobodan Ilic, Slobodan.Ilic@tum.de

Ivan Shugurov, ivan.shugurov@tum.de

Li Fu

Pengyuan Wang, pengyuan.wang@tum.de

Fabian Manhardt, fabian.manhardt@tum.de

Benjamin Busam, b.busam@tum.de

Location

Campus Garching