THI License Plate Dataset
In 2019 the first version of the THI license plate dataset (TLPD) was created to support the development of an algorithm capable of anonymize license plates from images collected using a dashboard camera. The images contained in the dataset are street views showing vehicles (parked and moving) as well as pedestrians. Today, the dataset contains more than 17,000 vehicle images and about 18,000 labelled license plates, that were captured at different angles, distances and under different illumination (day and night) and weather conditions (sunny weather, fog, snow). License plates from both passenger vehicles and motorcycles can be found in this dataset. The TLPD is suitable for training robust license plate detectors that should work in various scenarios, being at the moment one of the largest European Union license plate dataset.
Following the work initiated in 2019, the TDLP is also being extended to include more images of other types of vehicles (e.g. trucks). Furthermore, different artificial intelligence algorithms are being developed and finetuned. One of the actual developed models presents real time performance for simple GPUs, being capable of identify and blur with high accuracy license plates in different angels, lighting and weather conditions.