Vehicle number plate detection using TFOD API and OpenVINO

Sayak Paul

Sayak Paul

Kolkata, West Bengal

This project demonstrates the use of TensorFlow Object Detection API to automatically number plates (Indian) from vehicles. ...learn more

Project status: Published/In Market

Artificial Intelligence

Intel Technologies
OpenVINO, Movidius NCS

Code Samples [1]

Overview / Usage

This project demonstrates the use of TensorFlow Object Detection API to automatically number plates (Indian) from vehicles.

Dataset used: https://www.kaggle.com/dataturks/vehicle-number-plate-detection

To kick-start the model training process, I followed the steps from TensorFlow Object Detection API's official documentation: https://github.com/tensorflow/models/tree/master/research/object_detection. I used SSD_MobileNet_V1 architecture which was pretrained on the COCO dataset.

The main challenge was to prepare the data for TensorFlow Object Detection API. The dataset comes in YOLO dataset format, so a decent amount of effort has been put to prepare the dataset compatible for using with TensorFlow Object Detection API.

Methodology / Approach

Dataset used: https://www.kaggle.com/dataturks/vehicle-number-plate-detection

To kick-start the model training process, I followed the steps from TensorFlow Object Detection API's official documentation: https://github.com/tensorflow/models/tree/master/research/object_detection. I used SSD_MobileNet_V1 architecture which was pretrained on the COCO dataset.

The main challenge was to prepare the data for TensorFlow Object Detection API. The dataset comes in YOLO dataset format, so a decent amount of effort has been put to prepare the dataset compatible for using with TensorFlow Object Detection API. After the model was trained I used OpenVINO to further optimize the network so that I could run inference on my Neural Compute Stick.

Technologies Used

  • TensorFlow Object Detection API

  • Google Cloud Platform

  • OpenVINO

Repository

https://github.com/sayakpaul/Vehicle-Number-Plate-Detection

Collaborators

There are no people to show.

Comments (2)