TUYGUN

Brief

The second TEKNOFEST 2026 Aviation AI team Serhan captains — five of them, running Task 2 of the category: holding a UAV's position when GPS is denied. He designed the visual-odometry pipeline that estimates the aircraft's 3D trajectory from the camera alone and tracks it frame by frame against ground truth, and built the tooling the team's data flows through — LabelForge, a labelling and dataset-management interface written from scratch, and JSONtoYOLO, a converter that turns mixed label exports into YOLO training data in seconds. The team cleared the Preliminary Design Report at 87.5 and completed the Online Competition Simulation round. Alongside the engineering he runs the team: task planning, milestones and cross-team coordination.

What he built

• 01Visual-odometry pipeline estimating the UAV's 3D trajectory from the camera alone
• 02Trajectory tracked frame by frame against ground truth to score the localisation error
• 03LabelForge — a labelling and dataset-management GUI built from scratch to standardise QC on complex aerial data
• 04JSONtoYOLO — converts mixed JSON label exports straight into YOLO (.txt) format, cutting dataset turnaround to seconds
• 05System integration across five people, plus captaincy: task planning, milestones, cross-team coordination
• 06Cleared the Preliminary Design Report at 87.5, into the online simulation round

Stack

• Python
• OpenCV
• YOLOv11
• CustomTkinter
• NumPy
• Matplotlib