Shulou(Shulou.com)06/03 Report--

2020-04-01 18:52:05

Editor's note: this article is from the official account "academic headlines" (ID:SciTouTiao) on Wechat. Author: Xiao Ting, 36 Krypton is released by authorization.

Recently, New Science published a report: the US military is developing a portable face recognition device that can recognize targets a kilometer away.

This is the US Special Operations Command (SOCOM) ongoing "Advanced Warfare face recognition in remote Technology" project, which began in 2016 and demonstrated a working prototype in December last year, paving the way for a mass production version. At present, the study is under way, and SOCOM did not disclose when it will be available.

Remote face recognition system designed for UAV

The technology is originally designed for handheld devices and drones, and the target can even be identified without knowing or facing the camera. As a result, some human rights advocates have also expressed concern.

SOCOM's documentation also points out that the technology can be shared with law enforcement agencies.

Face recognition within one kilometer

According to reports, the device was developed by Secure Planet, a company based in Arlington, Virginia, which mainly produces remote face recognition devices based on digital SLR cameras and runs commercial face recognition software on standard laptops. The recognition range of these devices is generally about 300 meters, and extending the recognition distance is not as simple as adding a longer lens to the camera, because it increases the noise caused by lens vibration.

Atmospheric turbulence is also a key problem, because air, like a constantly changing mirror, has a thermo-light effect in warm weather. The challenge for the new ultra-long-distance recognition system is how to convert the captured image into an image that is clear enough to be used in passport processing software.

It is not clear to what extent Secure Planet has achieved this goal, but a team from the Australian Ministry of Defense recently developed an algorithm that can clean up atmospheric turbulence and help remote face recognition. Other researchers have tried to use convolution neural networks to convert a series of blurred images into a single clear image. Neural networks can also help extract pictures from complex environments. During passport control, the subject is motionless, looks at the camera, and has plenty of light.

In the wild, the face may be at an inappropriate angle, moving, poor light, or obscured by shadows, headscarves, sunglasses or other obstacles. Face recognition with neural network is essentially based on the method of face feature extraction, the main features are extracted by multi-layer network convolution and dimensionality reduction, and the feature extractor and classifier are automatically formed for the input of sample images. thus a model suitable for target detection and recognition is formed. The commonly used neural networks are convolution neural network (CNN), recurrent neural network (RNN) and depth neural network (DNN).

Similarly, machine learning systems can convert images taken in complex environments into images similar to passport photos. "in principle, such a technology can be used within a kilometer." "We haven't broken through the basic restrictions on remote face recognition, and we haven't pushed optics or algorithms to the limit," said Walter Scheirer of Notre Dame University in Indiana. However, Walter Scheirer doubts whether current technologies can achieve reliable identification in such a range. "I want to know what the imaging environment is like," he said. "did they do it on a sunny, sunny day?"

Looking at technology from the perspective of human rights

Rasha Abdul-Rahim of the human rights group Amnesty International called the technology disturbing. "under camouflage, technology will distinguish between combatants and non-combatants more accurately," Abdul-Rahim said. "We are seeing a dehumanizing initiative, and the threshold for the use of force is lowering."

Abdul-Rahim says this could be a step towards fully automated weapons that can find targets and attack them without effective human control. She called for an insightful public dialogue and called for the transparent use of the technology, especially how the face recognition system works.

"this technology should not be used at all until the government can prove that it complies with human rights law, especially if the algorithm will make a difference between life and death." Abdul-Rahim said.

Https://www.toutiao.com/a6810694245431640588/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.