cv-uav

Computer vision + YOLO pipeline for detecting edge-rich objects (triangles/hexagons) in a field and running YOLO only when the CV gate triggers.

What this does

• Captures real-time frames from a camera (including iPhone via USB if it shows up as a camera device)
• Runs grayscale + Canny edge detection
• Finds long straight lines via HoughLinesP
• Detects regular polygons (triangles, hexagons, etc.) from contours
• Computes edge density + angular variance
• Runs YOLO inference only when the CV gate indicates likely targets
• Draws bounding boxes and CV overlays on the laptop display

Setup

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Notes:

• ultralytics depends on PyTorch. If PyTorch is not installed, follow the official install instructions for your system.

Dataset layout (YOLOv8)

Create your dataset under data/ and update data/dataset.yaml with the right class names.

data/
  images/
    train/
    val/
  labels/
    train/
    val/
  dataset.yaml

Train YOLOv8

python train_yolo.py --data data/data.yaml --model yolov8n.pt --epochs 50 --imgsz 640

Checkpoints are saved under checkpoints/ and also copied to:

• checkpoints/best.pt
• checkpoints/last.pt

Run the live CV + YOLO pipeline

python main.py --source 0 --model checkpoints/best.pt

Useful options:

• --list-devices to probe available camera indices
• --source 1 or --source 2 if your iPhone shows up as another index
• --model checkpoints/best.pt to run your trained model

Using an iPhone over USB-C

• On macOS, iPhone can appear as a camera via Continuity Camera. Unlock the iPhone and open any camera-using app once so it registers, then use --list-devices to find the index.
• On other OSes, you may need a camera bridge app (e.g., Camo/DroidCam). Use the app’s provided device/stream URL as --source.

Project plan

See PROJECT_PLAN.md for the full, detailed plan and milestones.