Intelligent IoT – Nice to See You
In the next phase of IoT evolution, products are built for natural interaction with human beings. Similar to human-to-human interactions, the capability of knowing the human counterpart is key for machines to start a meaningful engagement.
Solutions providers have been using various forms of technologies, including motion sensors, RFID, bluetooth beacons, voice recognition and fingerprint scanners to help machines know their users. Among them, face technology has a distinct advantage of being the most natural and effortless for the user.
In A Practical Guide to Using Face Technology (Part 1), we looked at the basic pipeline of face recognition and analysis technology. We also talked about options that developers can leverage to incorporate face technology in their products. In this part, we will look at considerations to take into account when putting this technology to use.
Face Technology in Business – Increasing Reach
When talking about face recognition, people usually associate it with traditional surveillance and security monitoring applications. In recent years, face technology has spread to other domains. In this article, we will focus on face technology in business applications.
|Usage Categories||Examples||Technological Components|
|Photo/Video Library Management|
|Business Intelligence / Analytics|
Table 1: Application categories and examples
Table 1 lists some face technology use cases in several product categories. A number of those use cases are still in trial phase. Even though the technology is game-changing, it is also very sensitive.
On one hand, companies need to be proactive so that they will not be left behind. On the other hand, they should take time to develop the best practices to avoid missteps that could turn away customers. Complicating the issue further, the considerations are different in different contexts, countries and target demographics.
China – The Wild West of Face Recognition
At one end of the spectrum is China, where privacy regulations are much less strict than western countries. It is at the forefront of AI and face recognition research. Chinese companies are jumping on face recognition bandwagon at an astonishing speed.
The technology is widely used, from catching criminals to reuniting lost children with families. Besides, Chinese consumers seem to be more comfortable with face recognition technology in their daily lives. The environment allows companies to experiment with the concept on a much larger scale, like Alipay with face recognition.
Balancing Business’ and Users’ Interests
Contrary to China, the situation in western countries is quite different. Early adopters of face recognition often encounter some push back from privacy rights advocates. Though we are starting to see changes in consumer attitudes in certain usage scenarios, companies are generally more cautious in adopting the technology.
In transportation, air travelers are willing to accept the use of face recognition technology to improve boarding process (Figure 1). When compared to other inconveniences encountered in security checkpoint (e.g. full body pat down), face scanning is relatively lightweight.
However, shoppers are still not that enthusiastic about personalized advertisements like those shown in the movie Minority Report. The key is whether businesses can provide strong incentives to their patrons.
In applying face technology, we should have a clear picture of value propositions for both business and users. Use a product we developed for the fitness industry as an example. It offers personalized fitness coaching by seamlessly analyzing the user’s age and gender with a camera mounted on the exercise machine. It also can deliver personalized training programs to the user and track his/her workout history. All without requiring any ID, wireless beacon, or password. Using the camera, it can also monitor how the user exercises by detecting the cadence of running or cycling.
For gyms, the product allows them to engage with their individual members by better understanding the exercise patterns. For members, the product provides a very convenient way to track progress and incentives to keep exercising. With the value proposition in mind, it will help us to make intelligent trade-offs during the design process.
Then we need to pay attention to legal implications. In the US, the privacy laws are different in different states. The recent class action lawsuit against internet companies like Facebook, Google and Shutterfly in Illinois will set a precedent in how states regulate usage of biometrics data.
As a product vendor, you will need to understand the privacy statements of the technology provider, especially if face recognition technology is delivered through a cloud service platform. The information will need to be integrated into the privacy statements of your product. In addition, if the product is used by a business, you should work with them to develop their privacy policies too.
Even if it is legal, some end customers may not be always comfortable with face capture. Feedback from businesses and their customers should be used in guiding the design (e.g. The product allows users to easily opt-in or opt-out). As for the fitness product mentioned above, the initial user experience (UX) design choice we made was to have the user making a simple gesture to activate the face features.
Risk Analysis and Management
Next, we will need to analyze and handle failure scenarios. The first source of failures is the algorithm itself which could give incorrect results. Some technology providers publish the performance metrics (e.g. ROC curve, precision-recall curve, etc.) of their algorithms. That could be used for initial comparisons between providers.
Further evaluations should be done, using a data set that is similar to the actual use case. The results will be something like confusion matrix that presents important information like false positive rate and false negative rate.
Subsequently, trade-offs between precision and recall are made. Let’s say you decide to go for higher precision, i.e. maximizing the probability that a positive match result is indeed truly positive. The consequence is a higher rejection rate of true users, which will affect user experience. The product should be able to provide fallback for the users to easily identify themselves in case of initial rejection.
Other important risk factors include data security, cloud platform availability, and network connectivity. Those should all be taken into account.
Last but not least, there are a few product design and implementation choices that will affect the performance of the product:
- Camera Position – Place the camera at a position that can capture the face at the best distance and angle.
- Lighting – A well-lit environment with even lighting directed at the user’s face usually gives better results.
- UI – If available, the user interface should guide the user to look at the camera during the face scanning step.
- Pre-processing – There are some lightweight face pose detection libraries (e.g. OpenCV / DLIB) that developers can use to detect the presences of faces and measure the face poses. This can prevent sending poor face images to the cloud which will increase the cost and decrease the accuracy of results.
In this two part series, we walked through the basics of face technology and considerations in applying the technology in products. It is a core component for building human centric intelligent IoT products. At the same time, it is a fast evolving field where developers should continue to pay attention to changes in regulations, user behaviors, technology and market trends.