University of Twente Student Theses
Influence of Biological Cues on Monocular Depth Estimation
Groot Roessink, J. (2023) Influence of Biological Cues on Monocular Depth Estimation.
|
PDF
9MB |
| Abstract: | Monocular depth estimation (MDE) in computer vision is the process of estimating the distance to the camera for every pixel in a single 2D image, with that image being the only input data. In theory, this task is problematic since an infinite amount of 3D scenes can generate the same 2D image. Fortunately, for most real-world images there is little ambiguity for what its 3D scene should look like. In fact, current MDE methods exist that estimate depth that is close to the corresponding measured depth. Humans are also able to estimate depth from single-eye observations using prior knowledge about how certain cues in their observation (e.g. the size of a familiar object) relate to depth. In this work, these biological depth cues are extracted from an image and encoded into extra image channels. This new extended image is used as the input for an MDE method, which allows for studying the effects of explicit biological cues on M-DE. Observed effects in this study are substantial increases in estimation performance and efficiency in terms of training data. Furthermore, while the positive effects of cues on performance are most apparent for small amounts of data, the effects still remain substantial for much larger amounts of data. These results can be attributed to the fact that cue methods provide prior knowledge to an MDE method that it cannot learn from its training set. Since the cue method that contributes to the greatest performance increase uses unsupervised training, it provides a way of improving MDE using only unlabelled images. As a further contribution, a data set containing over 100K image-depth pairs is created using a realistic virtual environment. |
| Item Type: | Essay (Master) |
| Clients: | Info Support B.V., Veenendaal, NL |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science MSc (60300) |
| Link to this item: | https://purl.utwente.nl/essays/96632 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page