Detecting Road Intersections Automatically from Satellite Images using a Deep Learning Approach

Automatic detection of road intersections is an important task in various domains such as navigation, route planning, traffic prediction, and road network extraction. Road intersections range from simple three-way T-junctions (degree 3) to complex large-scale junctions with many branches. The location of intersections and their complexity is an important consideration in route planning, such as the requirement to avoid complex intersections on pedestrian journeys. This is relevant to vulnerable road users such as People with Blindness or Visually Impairment (PBVI) or children. Route planning applications, however, do not give information about the location or complexity of intersections as this information is not available at scale. As a first step to solving this problem, a mechanism for automatically mapping road intersection location and complexity is required, ideally using a globally available data source. In this paper, we propose a deep learning framework to automatically detect the location and degree of intersections from satellite images using convolutional neural networks. For this purpose, we labelled 7,342 \hl{Google maps images} from Washington, DC, USA to create a dataset. This dataset covers a region of 58.98 km$^{2}$ and has 7548 intersections. We then applied a recent object detection model (EfficientDet) to detect the location of intersections followed by a classification model (EfficientNet) to calculate their degree. Experiments based on the road network in Washington, DC, show that the accuracy of intersections detections is \hl{88.6\%, within 5 meters} Most of our predicted centre of the intersections (≈ 80\%) are within 2m of the ground truth centre. Additionally, our method detects the degree of intersections with an accuracy of 68.68\%. We have published an automation script to enable the reproduction of our dataset for other researchers. This work is beneficial not only for PBVI but for society overall.