Date of Award

Spring 2023

Document Type

Thesis

Terms of Use

© 2023 Lonnie D. Chien. This work is freely available courtesy of the author. It may be used under the terms of the Creative Commons Attribution 4.0 International (CC BY 4.0) license. For all other uses, please contact the copyright holder.

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Degree Name

Bachelor of Arts

Department

Engineering Department

First Advisor

Stephen Phillips

Abstract

Self-driving cars rely on sensors for their perception of the surrounding environment. Depth estimation provides crucial information for the control systems of autonomous vehicles, as avoiding collisions and accidents is impossible without knowledge of the 3-dimensional locations of other objects on the road. Monocular depth estimation, or depth estimation from a single camera, shows promise as a relatively inexpensive yet effective solution to the depth estimation task. Cameras, however, are limited by the image formation process in their ability to calculate 3-dimensional depth from 2-dimensional images. Traditional machine-learning models use images paired with ground-truth depth labels for training. The image-label pairs can be difficult to obtain, which calls for the development of unsupervised machine learning models. Sensor fusion, such as supplementing images with radar, also has the potential to improve monocular depth models. For the E90 project, we investigate the effectiveness of unsupervised machine learning models and sensor fusion in improving monocular depth estimation for self-driving cars.

Download

Included in

Engineering Commons

Share

COinS