RTSP Multi Camera Licence Plate Recognition using NVIDIA DeepStream

This project demonstrates a real-time license plate detection and recognition system built using NVIDIA DeepStream SDK. It uses multiple neural networks in a pipeline to detect vehicles, locate license plates, and perform OCR to read the plate numbers.

Sample Output

Visual Output

Terminal Output

Prerequisites

• NVIDIA DeepStream SDK 7.1
• Python 3.10
• Gst-python
• NVIDIA GPU with compute capability 6.0 or higher
• Ubuntu 20.04 or 22.04

Project Structure

Number_plate_recognition_Deepstream_OCR_NVIDIA/
├── models/
│   ├── lpdetect/              # License plate detection model
│   ├── lprecog/               # License plate recognition model  
│   └── trafficcamnet/         # Vehicle detection model
├── Results/                   # Output images and videos
├── nvinfer_custom_lpr_parser/ # Custom parsers for license plate recognition
├── deepstream_number_plate.py # Main application
├── dstest2_pgie_config.txt   # Primary detector config
├── dstest2_sgie1_config.txt  # Secondary detector config (LP detection)
├── dstest2_sgie2_config.txt  # Secondary detector config (LP recognition)
└── dstest2_tracker_config.txt # Object tracker config

Model Details

1. Primary Detector (pgie):

   • ResNet-18 based TrafficCamNet model
   • Detects vehicles, road signs, two-wheelers and persons
   • INT8 quantized for optimal performance

2. License Plate Detector (sgie1):

   • LPDNet model optimized for US license plates
   • Detects license plate regions on vehicles
   • INT8 quantized

3. License Plate Recognition (sgie2):

   • LPRNet model for OCR
   • Recognizes alphanumeric characters
   • FP16 precision

Pipeline Architecture

[File Source] -> [H264 Parser] -> [Decoder] -> [Stream Muxer] ->
[Primary Detector] -> [Tracker] -> [LP Detector] -> [LP Recognition] ->
[Video Converter] -> [On-Screen Display] -> [Video Sink]

Usage

1. Clone the repository:

git clone <repo-url>
cd Number_plate_recognition_Deepstream_OCR_NVIDIA

2. Run the application:

python3 deepstream_number_plate.py -i <path-to-video-file>

For example:

python3 deepstream_number_plate.py -i samples/sample1.mp4

Results

The application provides real-time visualization with:

• Bounding boxes around detected vehicles
• License plate regions highlighted
• Recognized license plate numbers displayed
• Frame statistics (FPS, object counts)

Sample results can be found in the Results/ folder.

Configuration

The behavior of each neural network can be customized through their respective config files:

• dstest2_pgie_config.txt: Vehicle detector settings
• dstest2_sgie1_config.txt: License plate detector settings
• dstest2_sgie2_config.txt: OCR model settings
• dstest2_tracker_config.txt: Object tracker parameters

License

Copyright (c) 2019-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved. Licensed under the Apache License, Version 2.0. See LICENSE file for details.

Acknowledgments

This project uses models and components from:

• NVIDIA DeepStream SDK
• NVIDIA TAO Toolkit
• NVIDIA TensorRT