What was accomplished

●       The YOLOv11 model was retrained with the augmented dataset from the previous round, improving mean average precision from approximately 78% to 91%.

●       Inference speed was benchmarked on the Raspberry Pi 5, confirming a processing rate sufficient for near-real-time detection during flight.

●       A batch of simulated flight-altitude test images was used to validate detection consistency across varying lighting and background conditions.

Challenges and solutions

The main challenge was maintaining detection accuracy while keeping inference fast enough to run onboard the Raspberry Pi 5 in real time. The team addressed this by using a smaller YOLOv8 model variant (YOLOv8n) optimized for edge devices, trading a small amount of accuracy for a significant gain in inference speed.

Media

Next steps plan

Begin building the ground station software setup around the dual-laptop MAVProxy configuration.