Towards HD Maps from Aerial Imagery: Robust Lane Marking Segmentation Using Country-Scale Imagery

Abstract

The upraise of autonomous driving technologies asks for maps characterized bya broad range of features and quality parameters, in contrast to traditional navigation maps which in most cases are enriched graph-based models. This paper tackles several uncertainties within the domain of HD Maps. The authors give an overview about the current state in extracting road features from aerial imagery for creating HD maps, before shifting the focus of the paper towards remote sensing technology. Possible data sources and their relevant parameters are listed. A random forest classifier is used, showing how these data can deliver HD Maps on a country-scale, meeting specific quality parameters.

Automatic Detection of Driving–Lane Geometry Based on Aerial Images and Existing Spatial Data

“…To exclude other objects with similar brightness, geometry characteristics (width, and length:width ratio) are used, because RSMs consist of narrow lines with a specified width. As Fischer et al (2018) suggest, roads (road axes) are used as buffers or limits to the area in which OBIA is applied, thus preventing misclassification of objects which are not located on roads. Nevertheless, redundant polygons such as cars, truck trailers, bright pavements or dust polygons (i.e.…”

“…However, computation in a raster environment is performance-intensive, and the use of topology and context rules is limited to a close neighbourhood of pixels. More recently, Fischer et al (2018) combined OpenStreet Map (OSM) road data with aerial imagery and detected RSMs using the Random Forest classifier and Gabor Filtering. However, their method was tested only on highways, and the authors anticipated inadequate results for other types of roads, notably in urban environments.…”

“…The main objective of the method proposed in this paper is to create a vector representation of driving lanes using aerial imagery and vector road axes, like Fischer et al (2018). However, the method is applicable to all types of road, including junctions, roundabouts, highway crossings and over-/underpasses.…”

Scenario Factory: Creating Safety-Critical Traffic Scenarios for Automated Vehicles

“…These works are capable of extracting high-level geometric information of road networks; however, lane-level information concerning motion planners of automated vehicles is not reconstructed. Promising works towards the reconstruction of road networks with lane-level detail from aerial images can be seen in [17], [18]. Similarly, while creating road networks for single lanes from OSM data is straightforward [19], creating those with lane-level detail is a more challenging problem [20].…”

Position estimation of mobile mapping imaging sensors using aerial imagery

“…The registration of the data sets is, however, a challenging task due to the overall differences between the image data sets. In current literature, registration approaches rely on aerial nadir images which entail the identification of mostly ground-based features that are salient in both data sets, such as road markings (Azimi et al, 2018;Berveglieri and Tommaselli, 2015;Fischer et al, 2018;Javanmardi et al, 2017). Despite the high accuracy which is potentially achievable with these methods, groundbased features may be occluded in the aerial nadir image or not even present in certain areas.…”