Compare Stereo Depth Methods for a Drone Inspection Startup
Overview
What this challenge is about.
You receive 500 calibrated stereo pairs from a turbine inspection plus sparse LiDAR ground truth on each pair. Implement (or wrap) three depth estimators: OpenCV Semi-Global Matching as baseline, a small learning-based model (e.g., HITNet), and a larger one (e.g., RAFT-Stereo). Measure depth accuracy (D1 metric and absolute depth error in centimeters at typical inspection distances of 2-8 m), runtime on a Jetson Orin (or a documented x86 proxy if hardware is unavailable), and edge-case behavior on shiny blade tips. Recommend one method and quantify the trade-off.
The Brief
What you'll do, and what you'll demonstrate.
Pick the best stereo depth method for blade inspection by trading off accuracy, edge-case robustness, and on-device runtime.
Earning criteria — what you'll demonstrate
- Implement and compare classical vs. learning-based stereo depth
- Quantify accuracy with standard metrics (D1, MAE) and edge-aware metrics
- Reason about the accuracy/latency/memory trade-off for edge deployment
- Defend a methodology choice in writing to a technical audience
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Career mappings coming soon.