Sightless Navigation using Absolute Monocular Depth Estimation

Absolute Monocular Depth Map of Stanford Men’s T&F Locker Room

This was my final project for BOTH of my two classes in Winter 2023. “CS 231A: Computer Vision: From 3D Reconstruction to Recognition” and “CS 238: Decision Making Under Uncertainty”.

It’s a smartphone app that connects to an endpoint that uses a monocular depth estimation model to allow the user to navigate around a room with only their smartphone camera.

In the end, while relative monocular depth estimation models are extremely powerful, their absolute counterparts leave a lot to be desired. While I was able to create a system allowing me to walk around a room with a blindfold on, various technical limitations made navigation slow and my precision less than surgical.

Read the paper here.

See the code here.