Some pedestrians walk down the sidewalk as if in a dream, absorbed by their smartphone screens, and drivers themselves are wont to become engrossed in a GPS app or text and lose track of the road. One group is harnessing the distracting smartphones themselves to keep everyone out of danger.
“We’ve created a system that can estimate if a pedestrian and a car are going to have an accident using only the smart devices that people carry,” says Kaustubh Dhondge, a graduate student at University of Missouri, Kansas City. “Instead of using a special communication mechanism, we’ve achieved this communication over Wi-Fi for the first time.”
Existing efforts for car-to-car or car-to-infrastructure interaction have relied on dedicated short-range wireless channels devoted to vehicle use, requiring specialized equipment. To make the technology useable without extra equipment for pedestrians and cars, the University of Missouri team turned to Wi-Fi.
Smartphones with Wi-Fi enabled send out short blasts of information at regular intervals as they search for their preferred wireless networks. For both pedestrians and drivers, WiFi-Honk squeezes extra information into those beacons about a phone’s position as determined by GPS, speed from the phone's accelerometer, and direction of travel from its gyroscope. Upon picking up signals from other units, WiFi-Honk uses a collision prediction algorithm to determine when a user needs to be warned of an approaching car or pedestrian.
Before now, a quick enough connection over Wi-Fi had been impossible: Normal Wi-Fi interaction requires the two devices to establish a connection, and Wi-Fi is unable to deliver data at speeds over 8-12 kilometers per hour (five or ten miles per hour). Wi-Fi beacons, on the other hand, can be captured at speeds up to 120 kph (75 mph). Phones send out these beacons every 100 milliseconds by default, but that timing can be quickened or slowed.
The group’s real-world experiments, as captured in this video (which won the Best Video Award at ACM Mobisys 2014, a mobile computing conference held last month in New Hampshire), tested a number of potential collision scenarios and simulated actual collisions to determine the necessary reaction time and alerts needed.