UWB Logo

Computer Science and Software Engineering Capstone Presentations

Summer Quarter

August 21, 2020

Anxin Liu

"Yoga Pose Detection System"

(UWB CSS Faculty Research)

Faculty Advisor: Dr. Min Chen

Abstract

This project builds a web app for the yoga pose detection using the following techniques: pose estimation technology, a deep learning model for human key point extraction, database building, and three-layer architecture. The back-end development of this project has two parts: yoga similarity calculator service and pose classification. The yoga similarity calculator service is already done. In order to meet the initial demand, a powerful machine learning image reorganization tool, OpenPose, is used to represent the first real-time multi-person system to jointly detect human body, hand, facial, and foot key points on single images. By calculating each angle on human body skeleton by the location data of key points, it is possible to recognize the pose on image. In this way, the system can compare the similarity of data of angles to find out the similarity of yoga pose on image. Also, a SQL database that stores pose name and its related angle data has been built on SQL Server. After users upload their pose images, the system calculates skeleton angle using OpenPose, then it finds the most similar data of angle in database and show the pose name as a result. The development of web app UI is still in progress. The Cloud service on Azure is used to build an wen app and store the user uploaded file. This project gives me an opportunity to get known about machine learning image reorganization. It might help me work on harder project in the future, such as human face recognition. By the interactive with database in back-end development, I deepened the understanding to the SQL database and learned how the database worked in a real project.

Updated August 19, 2020, 01:26