Have you ever wondered how to quickly verify your age? Without showing your ID or any other document. What tool could intuitively carry out this process? This is what we deal with at Codahead!

Nowadays, solutions such as self-service kiosks gaining popularity and automated point-of-sale (POS) systems finding their place in new niches, there’s a rising need for enhancing the user experience for customers using these machines. One area we explore at Codahead is age verification.

Accurately determining customer age opens up a whole variety of new possibilities for automated POS systems. For one it allows the kiosk systems to expand to new areas, where previously human supervision was essential, e.g., situations where customers need to be of legal age – alcohol or cigarette purchases (Supermarkets and Bars), lotteries (Lotteries), participation in gambling (Gambling and Gaming Industries).

Moreover, it makes age targeting possible. One can imagine a kiosk that displays a different interface for different age groups, e.g., more fun and engaging for teenagers, or easier to read and more streamlined for the elderly.

The most important part of the system that allows for such possibilities is the age verification process. For that we use an image acquired by the kiosk’s camera which is then analyzed by a deep convolutional neural network (InceptionV3). The network is trained with images of approx. 20 thousand persons with ages ranging from 0-116. 

If we analyze the problem of verifying if the customer is of legal age, we find out that the most difficult age group to recognize (the group in which the network output certainty is low) is 15-20 years old. This matches the real-life experience of clerks not being able to determine if a person about that age is of legal age. 

Taking that into consideration, for applications where determining legal age is crucial, we use biometric ID readers. This way it’s possible to securely obtain user age and the only challenge left is comparing the photo read from the ID with the picture from the camera. For this we use standard face recognition methods – Mobilenet networks converting face images to number vectors in such a way, that the euclidean distance of these networks is always low if the images represent the same person. We are using similar facial recognition methods that convert facial images into numerical vectors we are using in the SDK CodaFace library which is applicable in digital access solutions.

From the hardware point of view, the solution uses an HD camera to acquire customer images, a RaspberryPi computes module for on-site neural network predictions and an RFID reader to get data from customer IDs in cases requiring legal age check. Thanks to computations being run on-site (without the need for uploading images to an external server), we can ensure a fast and smooth user experience.

Self-service POS system popularity will continue to be a rising trend. In order to be ahead of the competition, one has to think outside the box and provide new ways to enhance user experience. The proposed solution opens up such possibilities by offering a fast and reliable way to determine customer age; information that can be used to provide a more personalized experience or open up the whole industry, where customers are required to be of legal age in order to make a purchase.

Age verification uses by Industry:

#artificialintelligence #computervision #ageverification #facerecognition #datascience #algoritms #bigdata