Brain4Cars system could spot driver errors before they happen
We keep hearing about systems designed to either alert drivers to impending collisions, to let them know that they've made a mistake (such as drifting out of their lane), or to tell them that they're getting tired. Brain4Cars, however, takes yet another approach. Created by scientists at Cornell and Stanford universities, it monitors drivers to determine when they're about to do something wrong, so it can warn them not to.
The system was developed by recording video of 10 drivers – along with the road in front of them – over a two-month period. Each person drove approximately 1,180 miles (1,899 km) on freeways and in city traffic.
When subsequently analyzing the footage, researchers used face detection and tracking software to see which head movements preceded which driving maneuvers (such as turns and lane changes).
They then assessed footage of another group of drivers, using the findings from the first group to guess what those people were about to do. The scientists' predictions were accurate 77.4 percent of the time, determining drivers' actions an average of 3.53 seconds before they were actually executed.
It is now hoped that a refined commercial version of Brain4Cars could be combined with GPS and map data about the roads that drivers were currently traveling. If a driver's head movements suggested that they were about to turn the wrong way up a one-way road (as an example), the system would alert them to the fact using audible and/or visual warnings.
The technology still has trouble when shadows pass across drivers' faces, when they turn their head to talk to passengers, or by variations in different drivers' head movements. To that end, the system may end up being augmented with features such as infrared and 3D cameras, along with other devices such as tactile sensors on the steering wheel, and cameras or pressure sensors to monitor drivers' feet on the pedals.
A demo of the system can be seen in the following video.