We can collect data using a smartphone camera or a Raspberry Pi camera and prepare the dataset by ourselves, or download existing images from the internet (that is, via Google, Bing, and so on) and prepare the dataset. Alternatively, we can use an existing open source dataset. For use case one, we have used a combination of both. We have downloaded an existing dataset on pothole images (one of the most common road faults) from and updated the dataset with more images from Google images. The open source dataset (PotDataset) for pothole recognition was published by Cranfield University, UK. The dataset includes images of pothole objects and non-pothole objects, including manholes, pavements, road markings, and shadows. The images were manually annotated and organized into the following folders:
- Manhole
- Pavement
- Pothole
- Road markings
- Shadow
