ROAD

Road Obstacle Aerial Dataset

Introduction

ROAD is a dataset of road that intended to use for road segmentation, it consists of 410 images of road captured from a drone at top-down angle. These images captured from various locations in Yogyakarta, Indonesia. This dataset is used for my undergraduate thesis about roadblock detection using semantic segmentation and path planning.

Description

The dataset captured using DJI Mavic Mini at 40m altitude. The captured data is in the form of 23 videos with 1920x1080 resolution. Each image is extracted from each video at 5 seconds interval and then cropped at the center with a square image ratio. Every image then manually labeled by humans.

There are 2 types of labels in the dataset, 3 class of segmentation mask (Road, Obstacle, and Background) and 2 class of classification label (Blocked, and Not Blocked).

Dataset and Label file size is 208 MB, Trimmed Raw Video 7.08 GB, and Raw Video 46 GB.