Vehicle Identification is a set of vehicle images to perform identification tasks of the brand, model and years of manufacturing.
This open database is built by collecting images from two different scenarios: images from web and images taken in real scenarios. Web nature images were collected from public websites and vehicle forums. The images taken in real settings were taken by authors in parking lots and public places.
All images have been obtained considering only the front view of the vehicles. The images show small variations in the angle at which they were taken, always ensuring that the main components of the front view of the vehicles are correctly appreciated. These variations allow the creation of more robust classification and identification models against these possible variations.
The images have been cropped by segmentation of the vehicles with the Faster R-CNN model. Once the segmentation is done, the smallest square containing the vehicle is obtained and then the vehicle plate is removed from the image.
|Original Image||Cropped Image||Image Without Plate|
The images have been labeled in classes, based on the brand, model and years of manufacture of the vehicles (e.g., Ford_Fiesta_2002-2005 or Nissan_Tiida_2006-Present). The present classes cover most of the best-selling vehicle models in Europe in the last ten years. This labeling for the images allows to organize the vehicle classes in a hierarchical structure, with three levels: brand, brand-model and brand-model-years, from lowest to highest degree of specification.
VehicleIdentification-1.0 contains a total of 3616 images belonging to 377 classes, 148 models and 9 vehicle brands (Audi, BMW, Citroen, Fiat, Opel, Peugeot, Renault, Seat and Volkswagen). The images have been taken from the front, and are labeled according to the make, model and years of manufacture of the vehicles.
|Dataset||Images||Vehicle Classes||Vehicle Models||Vehicle Makes|
The file structure of VehicleIdentification-1.0 is described below.