Creating Roadmaps in Aerial Images with GANs and Smoothing-based optimzation

Team members

Dragos Costea, Alina Marcu

Supervisors

Prof. dr. Marius Leordeanu

Prof. dr. Emil Slusanschi

Idea

Recognize roads and intersections at the higher semantic level of road graphs - with roads being edges that connect nodes.

Method

We present a method consisting of two stages:

• detect roads and intersections with a novel, dual-hop generative adversarial network (DH-GAN) that segments images at the level of pixels
• given the pixelwise road segmentation, we find its best covering road graph by applying a smoothing-based graph optimization procedure

Our approach is able to outperform recent published methods and baselines on a large dataset with European roads.

The raster detection of roads and intersections is depicted below

The road vectorization algorithm is described below:

Qualitative results of the road vectorization steps are shown below:

A - initial intersections (blue) and edges (green)

B - pruned edges (after removing collinear and overlapping points)

C - edges and intersections optimized with SBO (green edges with green vertices)

D - sampled leftover roads (green edges with red vertices) connected to the roads optimized with SBO

Qualitative results of extracted graphs are shown below:

Paper

UAVision Workshop, Camera-ready version, ICCV 2017

Cite

Costea, Dragos, et al. "Creating Roadmaps in Aerial Images with Generative Adversarial Networks and Smoothing-Based Optimization." ICCV Workshops. 2017.

@inproceedings{costea2017creating,

  title={Creating Roadmaps in Aerial Images with Generative Adversarial Networks and Smoothing-Based Optimization.},

  author={Costea, Dragos and Marcu, Alina and Leordeanu, Marius and Slusanschi, Emil},

  booktitle={ICCV Workshops},

  pages={2100--2109},

  year={2017}

}