Final Year Project

Project Title

Drone Image Stitching

Description

In this project, we developed a cost-effective alternative to satellite imaging by stitching high-resolution drone images into ultra-large panoramas. Using 150–200 images per scene at 4K–8K resolution, the pipeline applies classical feature detection algorithms (SIFT, SURF, ORB) alongside modern deep learning matchers (SuperGlue, LoFTR) to align and blend images seamlessly. Feature matching via BF Matcher and FLANN, combined with RANSAC-based homography estimation, ensures geometric accuracy, while multiband and feathering techniques produce smooth final outputs. The project benchmarks classical vs. deep learning approaches on accuracy and computational cost, offering insights into scalable, affordable aerial imaging solutions applicable to agriculture, urban planning, disaster response, and environmental monitoring.