High precision road reference data for autonomous driving applications in Open Drive Dr.-Ing. Gunnar Gräfe, 3D Mapping Solutions GmbH
Precise Digitalisation as a basis for OpenDrive data production Scanner data taken along Lombard Street in San Francisco, CA
Worldwide survey services... Worldwide kinematic survey services for Precise as-built digital road data production in OpenDrive Race track simulation High-end Surface models of (crg-projects) race tracks Proving grounds, test areas and special tracks with multiple resolutions Public roads
MoSES - Kinematic Road Mapping System Mobile survey system designed for road corridor mapping Calibrated sensor platform Object data acquisition based on: Precise position and orientation determination for the moving sensor platform Complementary sensor technology like cameras and laser scanners Post mission data processing 8 Cameras Up to 4 high Resolution Laser Scanners 1 Mill. Points per second Below 1 mm accuracy per point
Example for a kinematic 3D Laser Scanner Point Cloud Laser scanner point cloud (raw data) consists of: 3D coordinates for each point plus reflectivity of the target surface All vehicle movements are fully compensated. 1 million points per second per scanner lead to an exact digitalisation of the road surface with a point density of more than 3000 points per square meter.
Georeferenced, digital Image Documentation Documentation of road corridor and road surface Each image exactly georeferenced, calibrated and ready for photogrammetric object survey with our software 3D Road View. Multiple cameras (currently up to 10) with various possibilities of mounting and combination. The camera view for all cameras is flexible and may be adjusted by project.
Digital Road Data Production Objects are extracted from scanner data or photogrammetric camera data Typically longitudinal objects are extracted in scanner data and Point objects are measured using digital photogrammetry Object data extraction leads to 3D vector data, e.g. markings, curbs, guardrails, traffic lights, roadsigns, poles etc..
Digital Road Data Production: automatic data extraction Example: vector data extraction for automatic driving
Open Drive Projekt Example 1 German inner city 1151 axes with 54 km length 95 intersections 450 km of lanes 900 objects
Open Drive Projekt Example 1 Complex intersection modelling
Open Drive Projekt Example 1 Example: intersection modelling in OpenDrive
Open Drive Data Production example Example: highway intersection modelling in OpenDrive supported by imported Google Aerial images used for quality control
Open Drive Data Production example Example: Intersection modelling including complete 3D geometry, e.g. road surface crossfall, 3d road shape and 3D curbstone geometry.
Open Drive Projekt Example 2 German inner city 473 axes with 25 km length 1673 sections 71 intersections 165 km of lanes 2 km line objects 1050 signals 584 polygon objects (center islands, pedestrian walks etc.)
Open Drive Projekt Example 2 Intersection in OpenDrive including complete object information
Open Drive Data Production example Example: intersection in scanner data including all extracted geometry like curbs, markings, pedestrian walks etc.
Open Drive Data Production example Example: intersection in scanner data including Open Drive geometry like curbs, markings, pedestrian walks etc. Please note the traffic lights, for which each light is represented.
Open Drive Data connected with NDS navigation data Example: The blue lines indicate the existing NDS data, which was connected by NDS user data tags to the OpenDrive dataset.
Open Drive Projekt Example 3
Open Drive Projekt Example 3 Mapping parts of the Silicon Valley in OpenDrive as precise road reference data for autonomous driving development 1251 axes with 276 km length 317 intersections 2500 km of lanes 4500 signals 3900 point objects 220 km line objects
Open Drive Projekt Example 3
Open Drive Data texturing example ROD styles assigned by user data tags in OpenDrive
Open Drive Problems Open Drive is used to represent real road data for various purposes Typically the demand for precision is high (relative accuracy often required below 1 cm) Open CRG road surface models may be added Problems: Not all definitions in OpenDrive are tested for use with real road data. In detail time consuming problems occur. User requirements grow faster than the OpenDrive standard, so actually we see an inflation of user data tags. In detail there are some problems Definition of narrow curves (center may be within road boundaries) Center islands Parking places Stop lines non-transversal to the road axis Extended style guide also for inner-city layouts needed
Thank you for your attention! more information on: www.3d-mapping.de