Fun with Face ID and TrueDepth
Face recognition is standard on all the latest iPhones — and getting smarter
When Apple designed the iPhone X to be ‘all screen’, it created a couple of conundrums. There had to be somewhere to put the front–facing camera, so the notch was born. (Apple is one of several companies that have applied for patents on ways for cameras to shoot through or between pixels, so a truly all–screen phone should be possible one day.)
Meanwhile, the Home button had to be sacrificed, and with it Touch ID. Apple could simply have put the fingerprint sensor somewhere else, but instead it switched to a new method of biometric authentication using a sensor system, built into the notch, that it calls TrueDepth. It’s this hardware innovation that enables the iPhone X, XS, XS Max, and XR to support Face ID.
Facial recognition is a staple of science fiction and heist movies, but making it work in real life is hard. Simply picking out a face in the image captured by an ordinary camera would mean anyone who waved a photo of you at your iPhone would be as likely to get into it as you are. Earlier phones, including Samsung’s Galaxy Note 8, could be fooled in this way.
Instead of taking a picture, TrueDepth detects the three–dimensional shape of your face. The hardware, incorporated into the notch, is made for Apple by STMicroelectronics. On the right–hand side is a vertical–cavity surface–emitting laser (VCSEL). Just as an LED is a diode that emits light, a VCSEL is a diode that emits a laser beam.
The technology has been around for decades, but it’s new to the mass market, and the iPhone’s demand has vastly exceeded previous manufacturing capacity. In December 2017, Apple awarded $390m from its Advanced Manufacturing Fund to help optical component maker Finisar build a VCSEL factory in Texas, creating 500 jobs.
In front of the VCSEL is an optical filter that rapidly redirects its laser beam using tiny glass mirrors to project a grid of over 30,000 dots in a fraction of a second. The dots are invisible to the human eye, but a night–vision camera would see them
spring into action when the iPhone’s proximity sensor — third from the left in the notch — tells it a face is looming.
To the left again is another infrared source, a flood illuminator: basically infrared flash. With both this and the laser shining on your face, the TrueDepth infrared (IR) camera, the notch’s leftmost component, can capture both the dot pattern and a detailed image. As it’s IR, it works as well in daylight or darkness.
If the dots landed on a flat surface, they’d form a regular grid. On your face, the divergence of their positions can be used to calculate a 3D model.
It’s this 3D model that enables Face ID to recognize you. The optical image that’s captured simultaneously adds detail that’s used by iOS’s Attention Aware features, which recognize that your eyes are pointing at the device and avoid auto– dimming the screen while you’re looking at it, for example. By default, Face ID requires attention, so you have to make eye contact when presenting your face. This is to help avoid accidental unlocks and prevent someone grabbing your phone and unlocking it with your face. (The FBI, however, has already used a search warrant to demand that a suspect look at his iPhone X to unlock it.)
TrueDepth has much wider applications than Face ID. It’s also the basis of Animoji and Memoji. One day, incorporating a system similar to TrueDepth into a device’s rear camera array could enhance iOS’s augmented reality capabilities, which currently rely on the regular camera image and orientation sensing.
Apple has already added TrueDepth capabilities to the new iPad Pro. As for Macs, recent Apple patents describe both facial recognition (with automatic login) and using a depth sensor to allow macOS to be controlled by gestures in the air.
Five years ago, Apple acquired PrimeSense, which was behind the tech in Microsoft Kinect, but we have yet to see Apple pursue this kind of interaction on the desktop.
Visualizations of Face ID’s infrared flood image and dot mesh pattern.