Pages

2019-07-31

It Now Becomes Impossible to Lose Face in China: Faceless Face Recognition

As Chinese authorities seek cradle-to-grave surveillance in the world’s largest market for surveillance camera technology, AI algorithms take over where cameras cannot film faces.

A competitive arm’s race has been set off among Chinese companies. Similar to the “Sputnik shock” that set off the 1960s space race, this one seeks to generate the best surveillance technologies to meet the growing demand of police authorities in the country. Companies recently began to use software that recognizes people no longer by their face – which may be too easily fooled – but solely based on their body shape and gait. Algorithms are used whenever cameras cannot record faces. Gait recognition software is already being used by police on the streets of Beijing and Shanghai, but also, and potentially giving rise to more concern, for population control of Uighurs in the Western province of Xinjiang.

Virtually all surveillance products in use are made by Chinese companies in China. Foreign companies hardly play any role in monitoring the Chinese population as it has long exceeded a billion. In terms of gait recognition, startup Watrix has made a name for itself. CEO Huang Yongzhen said his system can identify people up to 50 meters away, even with their backs to the camera or with their face covered. Current European legislation banning facial or full-body veiling may be approaching obsolescence. Of course, the ancient trope of all Big Brothers in history surfaces again: “if you have nothing to hide, you have nothing to fear.”[1] Right – East Germany’s Ministry of State Security was built entirely on that sound principle. Similarly, China is less concerned with protecting citizens – and indeed surveillance can be shown to have potential benefits to that end – than it is with protecting the governing elite’s monopoly on power. Of course, human assets are becoming rapidly superfluous with surveillance technology based on pattern recognition, even though the computing power needed for gait recognition is far greater that what is required for face recognition, since it requires not one but a multitude of pictures to identify an unmistakable pattern. But watertight surveillance is essential to China’s massive experiment with ranking and monitoring citizens for purposes of social credits and demerits (very much worth a separate in-depth look – another day). The other pillar, of particular significance in China, is “trust your government.” Well… If Western societies that can freely elect and remove their governments cannot bring themselves to trust them, how trustworthy can be the Communist Party of China, an unelected monopoly that cannot be removed from power by lawful means and knows no term limits?

The increasingly sophisticated surveillance techniques have led human rights advocates to fear that Chinese people have very little privacy left. Robots resembling Star Wars’ "R2-D2" come equipped with dozens of sensors and cameras as well as red glowing "ears" and can identify individuals within a large crowd. Specialized police goggles can scan passersby and compare a person’s profile directly with a large database of fugitive suspects.

Of course, it is far from reasonable or even justifiable to view gait recognition as a “Chinese issue” that raises concern for individual liberties and human rights. The U.K., Japan, Israel and the U.S. Defense Information Systems Agency have worked on gait recognition software since more than a decade and certainly not without results, even if results have not been commercialized as they were by Watrix.

The Watrix software isolates an individual’s silhouette from video footage and analyzes its movement. Then it creates a model of individual gait. At this time, the algorithm is not capable of making positive identification of individuals in real time yet. It requires uploading of footage for purposes of analysis that can condense searching through an hour of surveillance footage to about ten minutes. However, no special cameras are required and footage from ordinary CCTV surveillance cams suffices to produce 94% accuracy. 

Purported uses of biometric recognition outside of social control purposes to maintain social stability and to manage society include the ability to spot people in distress, such as elderly individuals after a fall. But the closer one looks at claims that this technology can make life safer and more convenient, the more findings are reduced to fool-proof identification that cannot be derailed by limping, walking with splayed feet or hunching over because more extensive footage will invariably analyze all features of an entire body beyond commonly perceived visual traits. It would appear that gait is at least as unique to an individual as fingerprints or cornea patterns but features fewer abilities to escape detection. It is clear that Big Brother feeds on Big Data, and that the real danger lay in data preservation and storage, a fact that far exceeds legitimate purposes of near-real-time safety surveillance, such as monitoring a public space for elderly people after a fall.

Rewarding or punishing individuals for behavioral traits is hardly new and hardly a Chinese discovery. The key distinction is between criminal law – which, for good reason, has substantial barriers to system activation, such as burden of proof and presumption of innocence – and merely non-criminal if socially arguably less desirable choices. It is by that standard that China’s social credit ranking is likely to descend into an Orwellian dynamic.



[1] Daniel J. Solove, Nothing to Hide: The False Tradeoff Between Privacy and Security (2011).