D-ALPR: Drone-Based Automatic License Plate Recognition Within Restricted Parking Environment

Given its usefulness, License Plate Recognition using a drone is relatively less researched. Placing cameras everywhere is not possible. We can get images and video footage of unreachable places utilizing a drone with a camera. Publicly available datasets do not reasonably include samples with various challenging environmental conditions. So, we created a new dataset by flying the drone over open parking and basements of the premises of Nirma University and taking video footage in different conditions. After processing and labeling, the dataset is fed to the ResNet convolutional neural network to compare the accuracy of the existing approach with publicly available datasets and our dataset. We obtained an accuracy of 50–60% with our dataset, while the accuracy of the existing approaches is between 80–90% with publicly available datasets. It shows that there is a vast scope for improvement if conditions are challenging. The dataset created will be of immense value to the research community intending to work on this problem.