MORE | Spring 2024

Computer-Vision-Enabled Video Analysis for Motion Amount Quantification

Data icon, disabled. Four grey bars arranged like a vertical bar chart.

This work presents a novel approach to address the challenge of motion amount quantification for in-situ videos, focusing on the local and collective motion of body joints. Research approach considers a region-specific motion for particular body parts (e.g. only one hand). Given the task of motion amount quantification for videos, the project will consider relevant features from each frame and create a sequence of features, which is different from standard approaches involving calculating the optical flow of a region or using space-time neural networks to analyze the whole video. The novelty is that dealing with the features of body parts that fully contribute to the task and changes in the movement that will impact the performance of the task and result in fatigue. Pivotal features can be analyzed with various machine learning and statistical methods for the context of the task. This project proposes an integration of computer vision algorithms (i.e., deep neural networks for object recognition) and a statistical approach to accomplish this task.

Student researcher

Neel Hasmukhbhai Macwan

Robotics and autonomous systems

Hometown: Anand, Gujarat, India

Graduation date: Spring 2024