Background
For an automated vehicle (AV) to operate effectively it needs to precisely know its pose (position and orientation) locally and perhaps globally. This is called the “localization” problem. Without good localization, it is not possible to conduct precise maneuvers, avoid hitting objects, or even keep the vehicle reliably within a driving lane.
Ranger is an SwRI‐patented technology that allows sub‐decimeter precision in localizing a moving vehicle when driving over a pre‐mapped area. The system consists of a downward-facing camera with illuminators (lights) and an associated software suite. The performance of the Ranger system depends on the quality of the Ranger map, which is a registered and self-consistent set of features extracted from a series of images taken of the ground as the vehicle drives.
The goal of this project was to improve the Ranger map quality by designing a mapping system that allows a wider portion of the road surface to be covered in each pass. Another goal was to create an accurate metric reconstruction of the ground surface from the raw images. Such a system would help facilitate complex maneuvers such as lane changes using Ranger.
Approach
Several important steps were taken during this project. First, a novel hardware system was designed and fabricated that includes:
- Three cameras instead of one, which allows a much wider area to be mapped in each driving pass
- Electronics that synchronize the image acquisition of the three cameras with the pulsing of the illuminators
- The use of a global positioning system (GPS) time server that allows the camera images to be time stamped with a high degree of accuracy and precision
Software development was undertaken in parallel with the hardware design work:
- Creation of a software module to interface with a high-precision GPS time server
- Software to stitch the three camera streams together and register them to one another with respect to scale and warping
Accomplishments
We designed, fabricated, assembled, and tested a novel mapping rig that mounts to a vehicle hitch and uses three synchronized cameras to obtain a field-of-view that is greater than a lane width. Software was developed to support stitching the three image streams together and register them to one another in terms of position, scale, and capture time. Testing indicates that the system creates well-registered maps that are much wider than previous Ranger maps. The new maps have less perspective distortion and scale variance. These wider, higher-quality maps translate into improved localization of autonomous vehicles operating on the maps.