This is the final project for CSCI 1430 Computer Vision at Brown University. It was created in the Spring 2025 offering from Professor James Tompkin. This project was created by Kamya Raman, Max Smith-Stern, Stephanie Lin, Sue An Park.
The goal of our project was to recreate the popular game Just Dance using a three-step pipeline that takes in a depth image, isolates the person, classifies joints, and uses this joint vector to classify a pose. The current limitations of solutions to this problem are striking a balance between the amount of data and computing power we have, adapting to variations and noise in depth cameras, and mitigating the effects of having a long pipeline that is prone to errors.
Mapping 3D image data to 2D joint locations for pose classification is a computer vision technique that can be extended to solving other problems such as predicting pedestrian routes for autonomous vehicles and identifying improper movements for injury prevention. Our attempts to map 3D data to joint locations within the game platform are the beginning of a broader and more widely integrated technique.
Paper: Project Report
Poster: 