Publication Date

Fall 2021

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Engineering

Advisor

Mahima A. Suresh

Subject Areas

Computer engineering

Abstract

Micro-mobility has become a growing market that has altered transportation withincities. While helping people reach their destination efficiently while using fewer fossil fuels and resources, e-scooters lack tailored safety protocol. Using depth and ego-motion machine learning estimation through a live video stream, we hope to identify possible oncoming hazard for e-scooter users. To approach this problem, we tested methods that removed the pose network with a scaling transformation which was derived via linear regression. Our intuition was that training and inference will be faster with the removal of the pose network and was validated by the results. We also found that forward warping has good accuracy using the transformed ground truth egomotion over the relative egomotion from the pose network. This discovery can be used to predict if an object within the scene has a probability of colliding with the e-scooter user.

Download

Share

COinS