Images of roundabouts and junctions classified and labeled using the Labelme tool.

The dataset consists of 2000 PNG images (512 x 512 pixels) of roundabouts and junctions, with different dimensions and orientations. This dataset has been created to train a convolutional neural network for object detection (roundabouts and junctions) in various areas of the Community of Madrid.

This series of images has been obtained from the National Plan of Aerial Orthophotography (PNOA), and they will be classified and identified using the Labelme tool. Therefore, in addition to the .png format images, there are also .json files that contain the labels that have been created and identified in each of the downloaded images.

These data are classified into 7 types of classes: normal roundabouts (rotondaN), roundabouts split into junctions (rotondaEn), T-junctions without island (enlaceTsi), T-junctions with island (enlaceTci), U-turn junctions (enlaceTA), crossings, and lane change junctions (enlaceBI2).

The procedure followed is as follows:

1. Using the QGIS tool, a geopackage file was created to mark the different areas where an image was to be downloaded.2. Using a Python script, requests are made to the PNOA to download images.3. The various roundabouts and junctions obtained are labeled and classified.4. If more variability or data is needed, the process is repeated.

The shared dataset consists of a compressed file with the images and labels used for the completion of a final degree project.