Augmented Reality Tag Detection, Decoding & Imposition
(Feb 2020)
Project Overview
Developed a complete Augmented Reality pipeline for the ENPM673 Robot Perception course. The system detects custom AR tags in video streams, decodes their orientation/ID, and superimposes virtual content in 3D perspective.
Key Technical Implementations:
Custom Homography Estimation: Implemented the homography matrix calculation from scratch using Singular Value Decomposition (SVD) to map world coordinates to pixel coordinates, verifying accuracy against OpenCV's built-in functions.
Perspective Warping: Engineered a custom warping function to project 2D images (e.g., "Lena") onto detected tags, handling orientation checks to ensure correct alignment.
3D Projection: Deriving the Projection Matrix from the homography matrix to render a virtual 3D cube anchored to the tag's surface, requiring precise transformation of 3D world points into 2D homogeneous image coordinates.
Tag Decoding: Created logic to analyze the inner 2x2 grid of the tag, determining the Tag ID and compensating for camera rotation.
