When you see a crime or terrorist attack in London, a city that has one of the highest density municipal security camera systems on the planet, you must wonder why these crimes aren’t solved nearly instantly. The issue often isn’t camera coverage or the lack of an image, but that they have too much information to manage and parse.
This is one of the big problems that deep learning/machine learning systems are supposed to be able to address. This last week in China, NVIDIA showcased its solution, Metropolis, which is designed as a comprehensive answer to this problem. It is made up of DeepStream SDK, Jetson at the edge, Tesla P4 GPU accelerators and TensorRT for inferencing, and the DGX-1 AI supercomputer for training; NVIDIA also announced Asian partners Alibaba and Huawei, which will be helping bring the solution.
The demonstrated implementation was by Hikvision, one of the larger Chinese video surveillance companies, and claimed 90 percent accuracy in image recognition.
Let’s talk about why this is important for city-wide security and even security on large plant sites.
Where’s the Criminal?
Even with massive video camera deployments, security teams have their own version of Where’s Waldo. Theirs is called “where’s the criminal.” Today, you largely have to manually track an image from camera to camera to figure out where a criminal came from and to reconstruct the crime. Some solutions do have camera to camera tracking, but they are iffy and often the handoff between systems isn’t as clean as it needs to be.
However, in a city-wide deployment, where time is of the essence, it can take hours to days to wade through a significant amount of disjointed video footage, providing the criminal plenty of time to dispose of the evidence and skip town. Ideally, once a crime is identified, you should be able to immediately track the individual’s movements so you can identify their base of operations and/or locate where they are currently hiding.
To get there, we need systems that can scan massive amounts of video footage in real time, index the people in the footage, and then supply the video history of a suspect once that suspect’s face is identified.
This is what NVIDIA’s solution potentially will do. Once a suspect is identified, it will search for all the other images of that suspect and provide them in a time sequence so law enforcement can track that suspect’s movements prior to the crime and after it is committed. In addition, it should be able to flag that the suspect hasn’t left the site if they are still around and showcase habit patterns, which should not only help locate the suspect but provide information needed to assure that when the suspect is captured, there is minimal risk of collateral damage.
This capability not only will make suspected criminals easier to find, but it should reduce the risk to police officers during an arrest, and might make high-speed chases obsolete. If you can pick where the suspect will be, you just must go there to pick them up. The need to chase them on the road, along with the related risks to bystanders, would be significantly reduced.
It makes sense that one of the geographies aggressively deploying this technology would be China. Because of its massive population density, locating and subduing criminals is problematic and risky. This technology should reduce that risk substantially.
Wrapping Up: Site Security
While these systems are clearly designed for a substantial smart city type of deployment, they would also be viable for a large plant, education, government office, amusement park or military site, where tracking a potential attacker could be critical to either catch the suspect or prevent the attack in the first place. Just tracking a lost child at a Disney property can be a nightmare that this could prevent. It could likely also flag real-time threats like alligators and do a better job preventing that kind of attack as well.
We focus a lot on cybersecurity, but most of the physical risks are the good old-fashioned kind. Catching, or better yet, stopping someone before the crime is committed is becoming increasingly viable with deep learning and inferencing. These NVIDIA announcements showcase a huge step toward making our cities, plants, government offices and amusement parks safer and a near-term future where the bad guys can’t hide.
Rob Enderle is President and Principal Analyst of the Enderle Group, a forward-looking emerging technology advisory firm. With over 30 years’ experience in emerging technologies, he has provided regional and global companies with guidance in how to better target customer needs; create new business opportunities; anticipate technology changes; select vendors and products; and present their products in the best possible light. Rob covers the technology industry broadly. Before founding the Enderle Group, Rob was the Senior Research Fellow for Forrester Research and the Giga Information Group, and held senior positions at IBM and ROLM. Follow Rob on Twitter @enderle, on Facebook and on Google+