Safety system targets vehicle blind spot

4th October 2016

KU Leuven

Enaie Azambuja

0 0

In Europe, blind spot accidents involving trucks cause the loss of 1,300 lives each year. Unfortunately, existing blind spot mirrors and cameras can’t always prevent such accidents. “Blind spot mirrors are not always very effective,” says Kristof Van Beeck, doctoral student at the Embedded Artificially Intelligent Vision Engineering (EAVISE) research group at De Nayer Technology Campus in Sint-Katelijne-Waver.

“They never show the entire blind spot, and scraping past an overhanging branch with the truck may already be enough to displace the mirror. A blind spot camera, which shows the blind spot area on a screen using a wide-angle lens, is more accurate, but like the mirrors, they require the driver to stay focused. And it’s unavoidable that people sometimes get distracted.”

Van Beeck developed a safety system that automatically warns truck drivers when there’s someone in the blind spot. “The system uses complex algorithms to analyse the footage of blind spot cameras. By means of image recognition, the system detects cyclists and pedestrians. When someone shows up in the blind spot area, an alarm goes off in the driver’s cabin.”

To ensure sufficient safety, the image recognition has to be as quick and accurate as possible, says Van Beeck. “As the cameras show the entire blind spot area, the images on the monitor are distorted.

Relying on these images, our system has to detect objects as accurately as possible, thereby distinguishing between, say, a cyclist and a traffic sign. Furthermore, the system needs to respond immediately. High accuracy and speed tend to be mutually exclusive, but we have managed to combine the two nonetheless.”

Van Beeck and his research group also tested the system in practice. At the car technology centre at De Nayer Technology Campus, they simulated specific traffic situations involving a truck. “We have tested a wide range of blind spot situations,” says Van Beeck. “These simulations show that our system has a 95% accuracy.”

The researchers are currently talking to industrial partners to develop a commercial system for trucks. “But this requires further research. For instance, we have to fine-tune the system a little more and test it in actual traffic."

"We have to be sure that it continues to work during long drives and that it can handle weather situations involving rain and mist, which may render image recognition more difficult. But I am optimistic: I believe that a concrete application will be possible in the future.”

By Pieter-Jan Borgelioen. Translated by Katrien Bollen.