Portrait of Yakup Tekeli in a red jacketYakup Tekeli
Back to projects

2026 · Completed

Visual Localization for Drones

GPS denied UAV localization: estimates a drone's absolute position by matching its camera imagery to satellite maps (SIFT + LoFTR), then streams the position live to QGroundControl over MAVLink.

  • Python-QGC-Torch-MavLink-KalmanFilter-Georeferencing

Outcomes and learning

- Computer vision: feature matching, homography, and robust geometry estimation using RANSAC. - Experience with deep learning-based matching (PyTorch/Kornia) and GPU (Apple MPS) acceleration. - Georeferencing and map projections (Web Mercator, pixel↔WGS-84). - Smoothing of noisy measurements via state estimation (Kalman filter). - UAV communication protocols (MAVLink) and system integration.

Project document

Open in new tab