Ramanathapura Satyanarayana, Abhishek (2023) Enhancing depth estimation for Transparent objects. Master's Thesis / Essay, Artificial Intelligence.
|
Text
mAI_2023_AbhishekSatyanarayana.pdf Download (33MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (138kB) |
Abstract
Transparent objects are ubiquitous both in household and industrial environments. The recognition of transparent objects in the environment is very important for the ultimate aim of grasping such objects by autonomous robotic systems. And, for this, a grasping robotic system will require highquality RGB and depth images. However, the popular RGB-D cameras cannot provide accurate depth images for transparent objects due to their nature. In this research, we propose a novel architecture ResNet-50 + PSA model that can be used with the clearGrasp pipeline to estimate enhanced depth images from just the RGB image, especially for transparent objects. In the clearGrasp pipeline, the depth estimation task is divided into sub-tasks of transparent object segmentation, occlusion boundary segmentation, and surface normal estimation. The experiments demonstrate that the proposed ResNet-50 + PSA model is better than the DRN model as it achieved better performance on every sub-task as well as the final depth estimation task. It was also demonstrated that the clearGrasp pipeline with ResNet-50 + PSA model has better generalization on real-world novel objects.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Mohades Kasaei, S.H. and Valdenegro Toro, M.A. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 26 Oct 2023 10:54 |
| Last Modified: | 02 Nov 2023 14:57 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/31527 |
Actions (login required)
 |
View Item |