Vehicle Registration-Plate Detection With Machine Learning : a practical approach

Abstract

The thesis about vehicle registration plate detection, which is based on the problem of the existing parking system, PERCS. The parking system uses a third-party product (VRS) for vehicle recognition. But PERCS stakeholders decided to replace the VRS system and introduce their own mechanism to recognize vehicles. As a result, surveillance cameras (as IoT) were introduced for parking areas to capture vehicles and send them to another API service to identify vehicle’s details. In that case, all the captured images are sent to a vehicle recognition API. Sometimes, license plates of many vehicles are not visible in the images. Therefore, the vehicle recognition API returns empty result when it sends vehicles without license plates. PERCS management wanted to ignore vehicles without license plates for image recognition API, as it incurred additional costs at the end of month.

To address this issue, we decided to identify licence plate availability in the image before it sends to the vehicle recognition API. The task started with finding a suitable machine learning model. Also, required images to train the model are taken from PERCS system. This implementation hosts in MS Azure environment because PERCS systems are mostly hosted in Azure Web Apps.