Biometric system could turn vehicles into cash registers
Off-the-shelf dash cams can be augmented with biometric systems to turn vehicles into secure checkout registers for drivers and passengers using a phone app, according to researchers.
Vehicles are sold with similar built-in extras, the team claims, but they only recognize and act on behalf of the vehicle’s owner. The new prototype, called the DashCam Pay, can recognize multiple people at once with face and voice recognition and make payments for each of them separately at, for example, fast food restaurants.
A commercial version likely would be an Android app. The core system could be used to facilitate in-person transactions as well, according to the researchers.
In a new paper, the researchers write that people in a vehicle could address the DashCam Pay the way they might address Amazon’s Alexa home service, but in this case to order and pay for goods and services from their vehicle on the road.
A report just released by Goode Intelligence sees in-car payments as a key driver likely to push the automotive biometrics market to $560 million by 2026.
Food is the most obvious good to purchase, especially as sit-down eateries have started serving carry-out food to drivers during the pandemic.
But the researchers, two from Michigan State University and two from Visa Research, foresee the DashCam Pay also being used to pay for fuel and retail items. People would link their devices to the DashCam Pay to use it.
The devices would use face and voice biometrics to positively identify users as well as to place and accurately pay for orders.
In experiments, 120 minutes of combined video and audio were collected from 20 subjects in five vehicles at two sites.
A pretrained multi-task convolutional neural network used in experiments achieved a 99.1 percent true positive detection rate at a false positive rate of .01 percent. A pretrained FaceNet model showed a true positive ID rate of 98.9 percent at a false positive ID rate of .01 percent.
Mycroft AI precise, a trained trigger-phrase (“Hey, DashCam.”) detector, had a true positive rate of 98.2 percent at a false positive rate of 1 percent on test data.
The researchers used DeepSpeech, Mozilla’s open-source speech recognition algorithm, and after some massaging, its word error rate was cut to 3.65 percent.
For speaker recognition, an off-the-shelf system was enlisted, scoring a true positive ID rate of 98.4 percent at a false positive ID rate of .01 percent.
Privacy would be preserved using a biometric comparison protocol shared by the dash cam and mobile devices held by occupants of a vehicle. Security would be ensured by having the biometric data captured by the dash cam compared on the user’s device. The dash cam would only know whether a match was made or not.