Computer vision and sensor fusion have quickly become a key focus on the future of mobility. This month’s digital summit – ScaleUp 360° Auto Vision – brought together some of the industry’s brightest minds to discuss and dissect these important topics.
TomTom’s Alessio Colombo and Vincent Demuynck kicked off the first day of sessions with a look at how high-definition maps will aid the cars of tomorrow. Whether highlighting traffic signs, lane markings or other road attributes, this technology could prove to be invaluable. “The sensors in the vehicle recognize these lane markings,” said Demuynck, who serves as TomTom’s senior product marketeer. “What we provide in our HD map is information on the geometry of these markings.”
Demuynck added that the maps can “very accurately” depict the starting and ending point of each marking. This, he explained, “provides context for localization software to correlate real-world observations” with info from HD maps to determine where the car is and where it’s going, as well as its next appropriate course of action. TomTom’s HD maps already cover 380,000 kilometers of highways across the United States, Western Europe and Japan. But it’s not the only company looking at solutions.
When asked about the potential to reduce the need for HD maps using onboard sensors, Tudor Nicosevici, head of Continental’s ADAS Advanced Engineering department in Romania, proposed a different approach. “It is shown you can drive in a limited way without maps but maps gives you a much better understanding,” said Nicosevici. “The major, let’s say, concern regarding HD maps is they are very costly to develop.” Continental has developed a crowd-based solution that that can allow any car to contribute to the map, improving the accuracy for everyone.
Some within the industry have hoped that V2X will supersede the need for many of the maps and sensors currently being developed. Eric Schmidt, director of lead engineering and safety at TTTech Auto, isn’t convinced that is possible. “We will probably never be able to use [V2X] as the only source for decision making for autonomous driving,” said Schmidt. “I would consider it part of the overall architecture.”
Time for a (Virtual) Road Test
Before autonomous vehicles can hit the road, startups and OEMs alike need to test their driverless cars virtually. David Mear, business development manager EMEA at Vires, discussed the importance of placing buildings, obstacles, construction sites and other real-world elements into the environment to accurately create a simulated world. “You need reliable and realistic models, but your models need to be scalable as well, depending on what you want to test,” said Mear. For some use cases, Mear said that a simple model with extremely high performance may be warranted. For others, more accurate models with configurable performance might be required.
Sandeep Sovani, director of global automotive industry at ANSYS, concluded the two-day event by explaining why generic simulations are not enough – they must be tailored to automobiles. “You often see simulations done with gaming engines,” Sovani. “Gaming engines are designed for creating images that will be consumed by the human eye. Therefore they are not the best for doing simulations of autonomous vehicles because what the camera sees is different from what the human eye sees.”
In short, cameras may not be able to detect the finer details until they are very close, whereas the human eye might be able to recognize a traffic sign from far away. Physics simulation can help work around that. “We can actually test and train the perception software of a camera,” Sovani added while demonstrating a model that passes the images directly through the ECU of the camera. “The ECU thinks the images are coming from actual sensor of the camera, but instead they are coming from a simulation.”
About the author:
Louis Bedigian is an experienced journalist and contributor to various automotive trade publications. He is a dynamic writer, editor and communications specialist with expertise in the areas of journalism, promotional copy, PR, research and social networking.