(The original Chinese version of this article was written by MakerPro Editors and published on MakerPRO)
Amazon released a video of a cashier-free convenience store and caused discussions on the Internet. In the video, consumers purchased goods effortlessly, just put the goods in the shopping bag and made the payment without scanning the barcode. The video successfully made people have expectations of face recognition but resulted in lots of misunderstandings as well.
To know more about the operation of smart retail and the key issue— face recognition, MakerPRO interviewed the vice president of TCIT, Wan Yizhong, to talk about the related technologies, applications, and progression.
Two-Factor Authentication for Individual Recognition is Required
“In general, when it comes to face recognition, people may have the impression of opening a door by scanning the face or picking out individual faces in the crowd.” said Wan, “But in fact, it’s not the current technology that can achieve. Thus, the technology applied in cashier-free store and face recognition applications should include another authentication method.”
In the articles published by Amazon engineering team, we can find that the face recognition technology cannot and is not allowed to support the entire purchasing process. Therefore, consumers should download Amazon Go App first and scan the QR code while entering the cashier-free store. Then, no matter what you pick up and put back, the tiny RFID attached on the products will transmit the signals. So the products you choose will be recorded, and the payment will be deducted from your Amazon account.
In reality, face recognition can be mainly applied to analyze the number of people purchasing specific products. But in ID authentication, the system using face recognition should accompany another authentication, such as personal password, id card scanning, etc. The 2 ways to authenticate consumer identification is called “two-factor authentication.”
Recognition Technology Applied Widely: People-Flow Analysis & Security Monitoring
Wan said that recognition technology can be applied to smart retail but is not easily applied to the reality due to privacy issue. However, the image recognition technology— “scan then recognize” still can be applied widely. For example, this technology has been adopted in 7-11 convenient store for a long time to scan the crowd and then analyze the people-flow, age, gender, product viewing, duration time, and then record the data as the references for the next season purchasing. Also, the ad display system could apply the data to choose the most suitable message to display to you.
Besides people-flow analysis, face recognition can be applied to access control as well. For example, some Singapore government agencies adopt TCIT face scanning and BLE Beacon technology as the two-factor authentication as the electronic monitoring equipment to open/close door. With the technology, the accuracy and security of recognition can be even better than speedy immigration inspection. In addition, in a factory with numbers of workers, the image recognition technology could be applied to be the supplement to improve security authentication to monitor the people entering the factory.
The current image recognition technology is processed with machine learning. AI learning is similar to human learning so that it takes large quantities of data to analyze and compare the similar features to understand the objects. With the process of learning, image recognition being applied in daily life will need at least 5 pictures so that the technology could be used for unlock mobile phone and mobile payment, like fingerprint to unlock iPhone.
While being applied to recognize one person from multiple faces, such as access control, there might be a challenge. The larger the population of faces is, the worse the accuracy and effectiveness is. Since loading the large population will cause more loading, it will take more time to process. Therefore, one more authentication factor will be needed in recognition technology application.
In smart retail people-flow analysis, the more advanced solution is to apply the smart photographing with edge computing to extract each person’s characteristics in shooting, such as gender, age, duration time, etc. Then the data will be saved in database to reduce the time for saving video and transmitting data. It is the solution to strike a balance between “accuracy” and “effectiveness.”
Cross-platform and Hardware Production
Wan thinks that there are still lots of possibilities in the market and there’s room to improve in technology. For example, 3D camera could be applied to factuality detection to solve the problems of photo copying. Also, the face technology could cooperate with hardware companies to improve the compatibility and optimize the performance of the hardware devices.
Since more and more people are interested in face recognition technology, software companies have developed light-weight modules to support the edge computing need in the era of IoT. For example, for user-friendly issue, Wan said that TCIT face recognition system supports Windows 10 and Android system and separates face feature extraction and distinguishing from the front end and back end. That is to say, face features can be extracted from the device with limited computing capabilities, while facial recognition can be processed on the back-end powerful computer. The purpose is to allow more people to apply face recognition technologies.
To sum up, image recognition technologies will be applied in our daily life soon. No matter unlock smart phone, office check-in, or cashier-free retail, the applications can be expected. With the fast growth of new technologies, image recognition will be applied to face recognition and stock management very soon and will be the important breakthrough in the industries.