Running LPRnet.hef OCR on Hailo8 with Raspberry Pi 5 – Seeking Insights

Hello All,
I am New to hailo and am working with Hailo8 for the first time,. and am evaluating its use for the purpose given below.

I seek some insights and help for the same.

Objective:

I’m exploring the feasibility of running LPRnet.hef for OCR-based license plate recognition on Hailo8 with Raspberry Pi 5 as the host. However, I couldn’t find any example code or documentation on how to process an input image/video and extract OCR predictions from the hef model.

Key Questions:

  1. Model Execution: What’s the correct process to load and run LPRnet.hef on Hailo8?
  2. OCR Output Handling: Any guidance on extracting readable text (license plate numbers) from the inference output?
  3. Pipeline Integration: Best approach to integrate this with a video stream for real-time inference?
  4. Performance Considerations: Any known bottlenecks or optimizations for RPi 5 + Hailo8?

Looking for any starting points or references to get the hef running and extract OCR predictions. Appreciate any guidance!

Regards
Parth Bapaye

@parth
We published a guide on license plate recognition pipeline: A Comprehensive Guide to Building a License Plate Recognition (LPR) Systems. Please see if it helps. It covers most of your questions. However, please note that it does not use LPRnet as it can only detect digits and not letters. So, we use a custom model trained to recognize both digits and letters. Please let me know if you encounter any issues.