Two imperial lions, known as Shi in Chinese, stand guard in front of a traditional archway, or Paifang, outside DeepGlint’s offices in north-west Beijing. The guardian lions symbolise the kind of work DeepGlint does inside this picturesque campus set in Zhongguancun, known as the silicon valley of China.
The startup revolution has been booming over the years in China, but here, in Beijing, something has always been a Chinese version of something else from the west. DeepGlint, a tech startup that specialises in 3D computer vision and machine learning technologies, might be the start of something that is “disruptive” in the real sense of the word.
Bofei He, CEO of DeepGlint, says 12 guards monitor the 50,000 surveillance cameras at the Beijing airport, which is over 4,000 cameras per guard. It is difficult to keep track of every movement and, simultaneously, multiple camera feeds. The solution: Haomu Behavior Analysis System (HBAS). It is a camera system that can accurately reconstruct the three-dimensional scene, using a depth sensor, and track multiple targets and detect abnormalities.
“DeepGlint wants to enable computers to understand the world the way it is,” says He.
The 3D computer vision gives DeepGlint an edge over other surveillance systems using 2D vision as it doesn’t rely on the change of pixels inside the mapped area like the latter; thus avoiding false alarms even if someone simply moves around an restricted area and not enter the zone.
Two imperial guardian lions stand guard outside DeepGlint.
Banks in China have already begun using DeepGlint’s HBAS cameras to protect their ATM machines and customers from theft. Their cameras can track abnormal behaviour — running away from a cash machine or lingering for too long — and alert the authorities by sounding an alarm. The same technology can be applied inside airports after a few minor tweaks. The system can cast a wider net to detect those leaving their baggage unattended and people trying to evade security checks. The HBAS can also differentiate between inanimate, but moving, objects such as a wheelchair and a live person.
WATCH: How HBAS tracks people inside a train station
“Haomu Behavior Analysis System (HBAS) behaves similar to Microsoft’s Kinect. It can read and understand people’s actions like any motion gaming system,” says He.
DeepGlint are currently in talks with Tianamen Square municipality, whose cameras generate 1,800 TB of video recordings, to arm them with their technology.
“If you burn the 1,800TB video files on to DVDs, the pile will be taller than the Eiffel Tower,” says He.
Moreover, it will be difficult to go through hundreds and hundreds of hours of recordings to pin-point when exactly a crime was committed in the area.
Each camera, however, could cost as much as $2000 and using fifty thousand of them at an airport would leave a huge bill. But would you place a price on security post the Brussels airport bombing?
