The Pi I was running the backend on decided to retire early 🥲 A new one is on the way, so things will be back up soon! You can still explore the framework in the meantime. Click anywhere to continue :)

Reach out to me at:

siddharth22sharma@gmail.com
Master's Thesis — The Pennsylvania State University

Human Tracking In 3D Space Through Monocular Vision

Read it here

Overview

This is my Master's Thesis, I worked on it for a little over a year. The thesis focuses on the object tracking problem in computer vision. Specifically Tracking people/objects in 3D space without the use of special equipment for depth estimation. That is, just using images and videos from a regular smartphone camera, the program can accurately calculate the location of the subject in the 3D surroundings that it exists in.

Algorithm

Pipeline

  • Estimate scene geometry from monocular cues (vanishing lines / horizon).
  • Track object keypoints across frames in image space.
  • Lift 2D tracks to 3D via projective constraints + scale calibration.
  • Smooth & regularize the trajectory for stability.

Notes

  • Single‑camera input; no stereo/depth sensor required.
  • Output is a 3D polyline ready for interactive inspection.

Demo

Upload

Video preview

Interactive 3D trajectory viewer. Orbit, pan, and zoom.