License-Plate Detection (Vision) YOLOv8 plate detector

License-Plate Detection (Vision)

YOLOv8 plate detector + dual-engine OCR with video tracking and Romanian-format validation — single-image and video workflows.

- **YOLOv8 detection** with adjustable confidence threshold

- **Dual OCR engines** — PaddleOCR (primary, more accurate) + EasyOCR (fallback)

- **Video tracking** — ByteTrack assigns a stable ID to each physical plate across frames; per-track OCR aggregation picks the best reading per plate

- **Romanian format validation** — 42 county codes + post-OCR character-confusion fixes (`8↔B`, `0↔O`, `1↔I`, `5↔S`) and missing-county-letter restoration (e.g. `F 08 AGY` → `IF 08 AGY`)

- **Adaptive OCR preprocessing** — adaptive upscale to a target width + unsharp mask, tuned for small plates (drone-altitude pixel sizes)

- **Model selector** — sidebar dropdown auto-discovers every `.pt` in `model/` or `weights/`

- **CSV / JSON / image / video export** of detection results

- **Benchmark tooling** — measure OCR accuracy against plate width to know your effective minimum readable plate size