Print

HD Maps: Fine-grained Road Segmentation by Parsing Ground and Aerial Images

Creating detailed road maps

Currently used road maps are intended for the navigation of humans. The roads are stored as centerlines with connections to other roads without providing detailed information about the physical dimensions of the road, the number and width of lanes and the presence of sidewalks or parking spots. Creating maps containing also this information is important for infrastructure planning, creating realistic simulations over roads and supporting autonomous driving by improving the scene understanding and allowing precise self-localization.

Method

The paper in downloads presents an approach for augmenting the road maps with fine-grained categories by applying existing road maps, aerial images and ground images jointly. The proposed approach considers both the image evidence and the constraints on the road layout and the size of lanes. This allows robust estimation also in case when lane markings are not visible. The alignment of ground and aerial images is necessary as even when applying sophisticated GPS-IMU systems, registration errors can still occur. The registration of the ground images to the map can also function as precise self-localization of the vehicle within the road, an important task for path planning and safe driving.

Example for the output of the method. Left: aerial images. Right: ground images. Pink: road, yellow: parking spots, blue: sidewalk, red: the road detection in ground projected into the aerial image

Air-Ground-KITTI dataset

The dataset of annotated aerial and ground images used in the experiments is provided at downloads.

Figure 2: Examples of the dataset. Left: aerial and ground images. Right: the corresponding annotations (road, parking sports, sidewalk, building).

If you use the dataset, please cite our paper listed at links on the right.