Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
3D image-based Completion network (Semantic Scene Completion from dual Depth Images)
3D shape completion plays an important role in robotics and perception and has obtained significant development in recent years. Existing methods have shown impressive performance on various formats: voxel grids, meshes and point clouds. Object completion focus on the reconstruction of missing topologies and geometries of objects based on partial and incomplete observations coming from one or more 2D images and 3D depth information, for example, point clouds captured against occlusions, under weak illumination, or from limited viewpoints.
Keywords: 3D image-based Completion network, Semantic Scene Completion, Human-Robot Collaboration
please see attached PDF
please see attached PDF
Objectives of the project:
- Literature review on image-base completion networks
- Collect a dataset using the 3D cameras in a real industrial environment
- Train, evaluate, and compare multiple completion network models for different kind of objects in the scene
- Implement the best completion network on real-time situation
- Write an academic report summarizing and analyzing the results gathered in
previous tasks
Timeline:
This project can be a Master project or a bachelor thesis and the estimated duration is about 6 months.
This work takes place within the framework of the MINDLab (https://www.zhaw.ch/de/engineering/institute-zentrum/ims/mindlab/), in which our research group focusses on the following main topics:
- AI-based human action recognition for low-level activity primitives and high-level behavioral action types
- Real-time AI-based human-robot contact detection during physical human-robot collaboration
- Developing intelligent non-verbal communication for physical human-robot collaboration
- Human-robot safety enhancement through profound analysis of dummy-robot collision
Objectives of the project:
- Literature review on image-base completion networks
- Collect a dataset using the 3D cameras in a real industrial environment
- Train, evaluate, and compare multiple completion network models for different kind of objects in the scene
- Implement the best completion network on real-time situation
- Write an academic report summarizing and analyzing the results gathered in previous tasks
Timeline: This project can be a Master project or a bachelor thesis and the estimated duration is about 6 months.
This work takes place within the framework of the MINDLab (https://www.zhaw.ch/de/engineering/institute-zentrum/ims/mindlab/), in which our research group focusses on the following main topics:
- AI-based human action recognition for low-level activity primitives and high-level behavioral action types
- Real-time AI-based human-robot contact detection during physical human-robot collaboration
- Developing intelligent non-verbal communication for physical human-robot collaboration
- Human-robot safety enhancement through profound analysis of dummy-robot collision