Built for UTA's SCAI RoboMaster team, this project converts tournament recordings into a competition-oriented object-detection dataset. The lightweight nano model supports edge-ready inference, while C++ targeting logic turns detections into actionable yaw and pitch adjustments.
Engineering contributions
- Extract representative frames from RoboMaster tournament footage
- Manually annotate enemy robots and pressure plates as separate detection classes
- Train a lightweight YOLOv8n object-detection model for real-time, edge-ready inference
- Prepare the training workflow for expanded footage and difficult edge cases
- Translate pixel offsets between the aiming reticle and detected robots into C++ turret yaw and pitch commands
System path
01Match footage
02Frame extraction
03Annotation
04YOLO training
05Turret targeting
The repository is private, so this project is presented as a case study.