Subteam · 03

Perception.

We build the computer vision that lets the drone find targets with nobody at the controls.

About

Giving the drone
its eyes.

At competition the drone has to find the targets and map the search area on its own. We build the vision that does it: the camera, the detector, and the data it learns from.

Focus Areas

What we work on.

Target Detection & Classification

We train the detector that picks the competition targets out of aerial imagery.

Dataset Collection & Labeling

Two survey flights gave us 1,253 frames to label by hand. Synthetic scenes fill the gaps.

Camera & Calibration

We fitted a new lens, calibrated it on a ChArUco board, and are retraining the detector.

Survey Mapping

We stitch the map on the ground from photos the drone streams down as it flies.

Camera · 1920 x 1200 · global shutter Mannequin · 0.94 Tent · 0.91 Car · 0.88 3 targets located

Technology

At a glance.

Detector
YOLO11n, trained on Kaggle
On the aircraft
TensorRT FP16 on the Jetson Orin Nano Super
Camera
Arducam Jetvariety AR0234, global shutter, 6 mm lens
Targets
Mannequin, tent, car (SUAS 2026 classes)
Training data
Real flight frames in Label Studio plus Blender scenes
Mapping
Survey photos stitched on the ground

Meetings

When we meet.

General Meetings

Tuesdays 5:00 to 6:00 PM

Thursdays 6:00 to 7:00 PM

Slugworks BE-49

No experience needed. Come by and we will get you started.