Programm Details

Donnerstag, 24.01.2019   |   16:50 - 17:35 Uhr   |    KDo 1.7

Detection and Validation of Vignettes with Deep Learning and Computer Vision

To drive on an Austrian motorway, one must buy a toll sticker, called Vignette, and stick it on the windshield of the car. To check if everyone is driving with a valid Vignette, a hardware system containing cameras has been installed on the motorways that take continuously pictures of cars passing by. Normally, this system detects and validates the Vignettes automatically, but the hardware system struggles to detect all vignettes. The failed cases of non-detected vignettes are forwarded and processed by humans, that takes important resources away from doing more complex work.  To process these cases also automatically, the Liquid Studio built a proof of concept leveraging Google’s TensorFlow deep learning framework and computer vision algorithms from the OpenCV framework that first detects the vignettes, then classifies them into monthly, daily or yearly vignettes and in the end checks if the vignettes are currently valid. Eventually, around 60.5 % overall potential workload that has been done by humans could be saved using the developed PoC.



Natalie Faber
Accenture
AI Specialist

David Sharma
Accenture
AI Developer