Samer-Gassouma/Tunisian-Licence-plate-Detection-ReaderOCR

# WaslaVision — Multi-Camera Tunisian License Plate Recognition Real-time license plate recognition pipeline for the **Nabeul Bus Station**. Processes 8 Hikvision RTSP cameras simultaneously at 30 fps using YOLOv8s TensorRT batch inference + PaddleOCR + gRPC output. ## Station Layout ``` 4 physical lanes × 2 cameras each (LEFT / RIGHT) = 8 cameras → 8 destinations Camera 1L → Destination 1 Camera 1R → Destination 2 Camera 2L → Destination 3 Camera 2R → Destination 4 Camera 3L → Destination 5 Camera 3R → Destination 6 Camera 4L → Destination 7 Camera 4R → Destination 8 ``` ## Architecture ``` ┌─────────────────────────────────────────────────────────────────┐ │ main.py (entry point) │ ├─────────────────────────────────────────────────────────────────┤ │ │ │ 8 × CameraStream threads (producers) │ │ │ RTSP → FFmpeg → decode → push to shared queue │ │ └──────────────────┬─────────────────────────────────────── │ │ ▼ queue.Queue │ │ ┌──────────────┐ │ │ │ BatchProcessor│ (single consumer thread) │ │ │ YOLO TensorRT │ batch=8 inference │ │ └──────┬───────┘ │ │ │ │ │ ┌────────────┼────────────┐ │ │ ▼ ▼ ▼ │ │ Per-camera pipeline (×8): │ │ VirtualZoneFilter → StabilityCheck → Crop → OCR → Voter │ │ │ │ │ ▼ │ │ gRPC → Go queue-service │ └────────────────────── …