Developing fast and unbiased computer vision algorithms – UROP Spring Symposium 2021

Developing fast and unbiased computer vision algorithms

Claire Wan


Pronouns: she,her,hers

Research Mentor(s): Carol Flannagan, Research Professor
Research Mentor School/College/Department: University of Michigan Transportation Research Institute, College of Engineering
Presentation Date: Thursday, April 22, 2021
Session: Session 1 (10am-10:50am)
Breakout Room: Room 16
Presenter: 8

Event Link


Images and videos are a common way to easily store data for research. However, turning these into usable and analyze data is much more time consuming and costly. In order to help research teams save time and money, DEVIATE is developing a computer vision algorithm to label image and video data for them. An essential part of the development process is to consider and mitigate any possible bias. In order to reduce bias, DEVIATE is using human coders to label the training and testing data. In addition, the pool of data represents a diversity of skin tones as well as different light levels. It is especially important to reduce bias since this product will be used by other research groups. If the algorithm was biased, it would affect countless other research studies as well, so it is essential that DEVIATE is able to minimize any harm coming from its product. As this is still an ongoing project, it is expected that the final product will be an unbiased comprehensive computer vision algorithm that can help research teams label data stored in images and video.

Authors: Claire Wan, Carol Flannagan
Research Method: Data Collection and Analysis

lsa logoum logo