Date of Award

5-30-2019

Document Type

Thesis (Undergraduate)

Department

Department of Computer Science

First Advisor

Lorie Loeb

Abstract

Annotation and labeling is a critical component of computer vision. However, completed manually, this process is time and cost-intensive. In particular, video annotation is particularly arduous in terms of manual annotation and therefore is additionally costly. Because videos are often annotated frame by frame, making little use of the fact that the data between any two consecutive frames are closely related, the process of completing a single video annotation is the equivalent of the cumulative work of annotating and labeling an equal number of distinct images. For certain applications of video annotation, we can leverage assumptions about the objects in the video that allows us to most efficiently utilize the similarity between the frames of a video. In this paper, I describe an analysis of a new software that implements a binary search algorithm for the use in video annotation for the social sciences. I present a specific case study that describes and analyzes the usage and efficacy of this software for tracking individuals in a social gathering for psychological research and discuss how the software may be used in other similar applications in the future.

Comments

Originally posted in the Dartmouth College Computer Science Technical Report Series, number TR2019-869.

Share

COinS