AI helps vehicles to find their way through roadworks
Building sites on highways are a challenge for automated vehicles: Narrow lanes tighten traffic lanes and cause congestions. Stressed drivers frequently react in an unsafe way which often leads to accidents. The sensor systems and algorithms of automated driving systems are overwhelmed by the complex situation where lane markings overlap, limiting beacons and traffic cones are difficult to identify by the sensors. Road signs contain diverging information on allowed speeds or the course of the tracks.
The researchers from Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS) in St. Augustin (Germany) are attacking these problems with Artificial Intelligence. “Our technology enables systems to read and understand road signs of the kind described with high accuracy,” said Stefan Eickeler who is responsible for object recognition at Fraunhofer IAIS. The data are acquired and subsequently they are processed at the semantic level to make sure the system can “understand” them at the contentual level. At this level, they then are made available to further processing instances in the vehicle. “With Deep learning we teach the software to faster and more efficiently identify relevant patterns,” Eickeler said.
With this method, navigation system and driver assistance system can collaborate in the future to correctly diverging exit markings on motorways, distances between cars can be adjusted more exactly and the speed can be set in a timely manner.
The equipment under development will be based on an automotive camera that currently is shooting 20 to 25 frames per second. The system can analyze the images on the fly, identify relevant information regarding information signs, traffic lane markings and LED-powered traffic signs (which is not a trivial task since the LED segments are powered at different instances). A future vision: The camera will establish a position as the primary interface for the autopilot, making multiple sensor types redundant and dispensable.
The current works are part of the research project AutoConstruct, launched in December 2016 and in part funded by the German government. The goal of the project is devising a series-ready, cost-optimized camera as a sensor for automated driving. In this context, Fraunhofer IAIS has assumed the task of developing the related image processing software.
Related articles:
Embedded ADAS Algorithm Optimization with High-Performance DSP IP and CV Software Library
Stepping into next generation ADAS multi-camera architectures
More information: https://www.iais.fraunhofer.de/en.html
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
