Building a Smart ALPR Dashboard: Real-Time Vehicle Tracking with Flask and YOLOv8
Back to Blog
Featured

Building a Smart ALPR Dashboard: Real-Time Vehicle Tracking with Flask and YOLOv8

Computer Vision Project

Waqas Kamran MarwatResearch Assistant
February 26, 20260 min read3 views

Automated License Plate Recognition (ALPR) is no longer reserved for high-budget municipal infrastructure. With the convergence of YOLOv8 for object detection and PaddleOCR for text extraction, developers can now deploy sophisticated tracking systems on standard hardware.

This guide explores a Flask-based ALPR system designed for real-time monitoring, direction estimation, and historical data logging.

The Architecture: How It Works

The system operates on a multi-stage computer vision pipeline that transforms raw video frames into actionable metadata.

  1. Vehicle Detection: A YOLOv8 model identifies vehicle bounding boxes.

  2. Plate Localization: A secondary, specialized model zooms in to find the license plate within the vehicle's coordinates.

  3. OCR Processing: PaddleOCR converts the visual plate into a string of text.

  4. Directional Logic: By tracking the centroid movement across a static "trigger line," the system determines if a vehicle is Entering (IN) or Exiting (OUT).

  5. Persistence: Data is cached in a CSV for lightweight logging, while snapshots are saved as visual evidence.

Core Features

  • Dual-Model Precision: Using separate weights for vehicles (car_model.pt) and plates (plate_model.pt) reduces false positives and improves accuracy in crowded frames.

  • Live Web Dashboard: A Flask frontend allows users to input RTSP streams (for IP cameras) or local webcam IDs to start monitoring instantly.

  • Intelligent Search: A built-in historical log viewer to filter detections by plate number, date ranges, or movement direction.

  • Automated Evidence: Every detection triggers an image save to static/detections/, ensuring you have visual proof to back up the OCR data.

Technical Implementation

Prerequisites & Setup

You will need Python 3.8+ and a few heavy-hitting libraries. Install them via terminal:

Bash

pip install flask pandas opencv-python ultralytics paddlepaddle paddleocr

Directory Structure

A clean structure is vital for Flask to serve static assets and templates correctly:

  • app.py: The Flask server handling routing.

  • pipeline.py: The "brain" containing the CV logic.

  • templates/: UI files (index.html, search.html).

  • static/detections/: Storage for captured plate images.

Critical Evaluation & Assumptions

While this architecture is robust for a "Smart Dashboard," there are several technical hurdles and potential biases to keep in mind:

  • The "Static Line" Assumption: Directional logic based on a static line assumes a fixed camera angle. If the camera vibrates or is repositioned, the "IN/OUT" logic may invert or fail.

  • Hardware Bottlenecks: Running two YOLO models plus OCR in real-time is computationally expensive. On a standard CPU, you may experience significant latency (frame lag). To achieve "Real-Time" status, CUDA-enabled GPU acceleration is practically mandatory.

  • OCR Sensitivity: PaddleOCR is highly accurate but can struggle with skewed angles or poor lighting. A "Confidence Threshold" should be implemented to discard low-certainty reads before they hit your database.