Instruction file imported from estebanfoucher/Sail-CV (.cursor/rules/tracking-module.mdc). Copyright stays with the author.

Tell-Tales Tracking Module

Overview

PyTorch-based tell-tale tracker for aerodynamics monitoring. Uses fine-tuned RT-DETR detector + custom ByteTracker.

Key Components

• detector.py: RT-DETR model wrapper for tell-tale detection
• tracker.py + tracker_utils/byte_tracker.py: tracking pipeline
• pipeline.py: Full processing pipeline (detect + track + PCA analysis + classification)
• crop_module/: PCA-based crop analysis, background subtraction (OpenCV MOG2, VPI)
• models/: Pydantic data models (Detection, Track, BoundingBox, Image, Layout, etc.)
• classifyer/: Tell-tale state classification
• video.py: VideoReader, FFmpegVideoWriter (simpler version, no stereo)
• streamer.py: Frame streaming for real-time processing

Model Checkpoints

• Located at checkpoints/ (project root)
• rt-detr.pt: RT-DETR fine-tuned detector
• yolo-s.pt: YOLO-S alternative detector

Configuration

• YAML parameter files in parameters/
• Layout definitions in assets/tracking/layouts/

Testing

• Tests in tests/tracking/
• Run with: make test-tracking
• Fixtures in fixtures/ (project root)