You are driving toward a busy London intersection but don’t see a pedestrian step onto the crosswalk in front of your car.
Computer Science associate professor Steven Beauchemin is in the driver’s seat of a new research study looking to make vehicles safe and reduce driver error. Along with colleagues at Western and Laval University, he has instrumented a vehicle that is able to interpret the driving environment and the driver’s level of concentration.
You don’t slow down, but suddenly your car brakes on its own.
This may sound like a car of the future, but researchers in the Department of Computer Science at The University of Western Ontario are partway there in making this advance a reality.
The human driver is often the weak link in our increasingly complex and congested transportation systems.
However researchers believe technology can give drivers an edge, helping them to make better, faster decisions to avoid hazards. These true smart cars – when fully developed – will ‘see’ the conditions around them, assess a driver’s response against an appropriate response, and take action if needed.
Computer Science associate professor Steven Beauchemin, professor Mike Bauer and a team of graduate students have partnered with Laval University researchers Denis Laurendeau, an electrical and computer engineering professor, and Normand Teasdale, a medical school professor, on the Road Lab advanced driving assistance system.
“The idea is to predict what the driver is going to do given the traffic environment around the driver or around the car, and … the onboard computer reads off the car network data such as speed, state of pedals – brakes in particular – steering wheel orientation, etc. in real time,” says Beauchemin. Beauchemin’s own car, which he donated to the project, has become a mobile lab for the system.
Four cameras are installed across the front windshield to capture short- and long-range images of what the car ‘sees’ from the hood to 180 metres ahead. There are also rear and side cameras, providing a 360-degree view. Facing the driver will be two cameras with an infrared light reading the driver’s eyes movements. Is he drowsy? Is she inattentive?
Data from the car’s diagnostic system, cameras, and a global positioning system (GPS) is fed to a super computer in the back seat to be analyzed. With this information, it is possible to determine how the driver should react in a situation, says Beauchemin.
The researchers are working on software to process the immense amount of information acquired from the fully-instrumented car and turn it into a measured response.
The goal is a system that “if the driver is not doing what he should be doing, then the system can take action,” he says.
Development of a Knight Rider-like vehicle able to temporarily drive independently using artificial intelligence is on the horizon. At this stage, the research team is focused on acquiring a better understanding of driver behaviour and developing computer response models.
“The research question is this: can we develop cognitive models of driver behaviour that are accurate enough or that work sufficiently well enough so that we can compare that with actual driver behaviour and attempt to correct it?,” says Beauchemin.
“Ultimately if you can make a car crash less, then you can make it much lighter because it doesn’t have to meet crash safety regulations,” he says, noting this would improve fuel economy and passenger safety.
Vehicles with artificial intelligence could help older drivers retain their driving permits even while reducing accidents. If all cars had “crash avoidance,” insurance rates should plummet, says Bauer.
This type of research is essential to eventual development of a fully-automated vehicle. Bauer foresees this technology having a huge impact on highway traffic.
More vehicles could fit on highways because they could safely travel closer together, reducing the need for highway widenings. Vehicles would communicate with each other when braking, for instance, so all cars would brake simultaneously, reducing fender-benders.
The goal is to eventually develop a car that would prevent or warn the driver of an upcoming hazard or event requiring an action, or correct mistakes by taking over the car’s operation for a short period.
“We are in a position to tell exactly in real time at a frequency of 60 times a second where the driver is looking so we can map that onto the environment that is detected by the camera looking outside the car. We can tell if the driver has seen or not seen a street sign, a pedestrian, or another car,” says Beauchemin.
Researchers hope to develop systems that warn a driver of an undetected issue, such as a pedestrian crossing the road.
“It’s going to keep you safe,” says Beauchemin. “It will make sure you are performing the right driving actions at any moment.”