DIY License Plate Reader #RaspberryPI #MachineLearning #Yolo3 #Keras @robertlchiriac

Robert Lucian Chiriac’s article describes a DIY device using ML to predict license plates in real-time.

 

Earlier this year Rober Lucian Chiriac wrote about creating a DIY device that can recognize and read license plates. The plate reader utilizes a combination of on-device and cloud computing to work in real-time. The hardware for the project includes a Raspberry Pi, Pi camera, 4G antenna, and a GPS antenna. These items are housed in a DIY 3D printed case designed by @robertlchiriac.

Once the hardware was assembled, @robertlchiriac focused on training an ML pipeline to process video from the Pi camera into license plate predictions. This task was broken down into two main chunks, recognizing the license plate, and predicting the characters on the plate.

Yolo3 was chosen for the license plate recognition task and was fine-tuned on a small dataset of 534 images annotated by VOTT. The final model is available here and the dataset is available here.

keras-OCR was used to detect the characters on the license plates.

Once the hardware and the model were ready to go, Cortex was used to pull the two together to get the predictions in real-time. If you’re interested in seeing it in action, check it out on YouTube. If you’re interested in more projects like this checkout @robertlchiriac’s website.

 

 

Written by Rebecca Minich, Product Analyst, Data Science at Google. Opinions expressed are solely my own and do not express the views or opinions of my employer.