The Crunch: Using powerful technology once reserved for making crystal clear gaming graphics, NVIDIA is now turning cameras into powerful data machines for retailers. By leveraging graphic processing units (GPUs) that work at an exponentially faster rate than traditional central processing units (CPUs), a massive amount of information can be analyzed in a fraction of the time. That processing power allows for deep learning and smart neural networks to be used to identify customer trends in retail. And the NVIDIA Metropolis platform is working to use those same commercial insights to make smart cities that are efficient and safe, enhancing the power of a vast video network that already exists.
As video games have become hyper-realistic and we have the ability to navigate seemingly infinite worlds, the technology that powers those visuals has grown. A big leap came in 1999 when NVIDIA introduced its Graphics Processing Unit (GPU) which is a single chip processor capable of processing a minimum of 10 million polygons per second.
The immense power dwarfed that of central processing units (CPUs) and gave gamers a faster, more visually stunning experience. The difference between the strength of the CPU and the GPU is probably best summed up by a demonstration from Adam Savage and Jamie Hyneman of the show Mythbusters. The duo created a paintball-shooting robot representing a CPU that paints a smiley face one paintball at a time — taking minutes. Then they unveiled a machine representing a GPU that shot 1,100 paintballs simultaneously, painting the Mona Lisa in just 80 milliseconds.
In 2008, NVIDIA determined that the capabilities of the GPU could be used for other purposes, allowing exponentially faster processing of data beyond just visuals.
“We found that the GPU, which was thought to be just for graphics, could be leveraged for more general purposes,” Adam Scraba, Global Business Development Lead at NVIDIA, told us. “A CPU could typically do sequential if-then processes, but the GPU was built to process a lot of data in parallel, which is the basis for a lot of scientific computing algorithms.”
So, NVIDIA created the general-purpose GPU (GPGPU), and it was evident very early on that the massive computing power could have a transformational effect on business, in particular with the emergence of machine learning and artificial intelligence.
Since then, NVIDIA has been harnessing the power of the GPU to create neural networks that can learn from the data they collect, making insights faster and more actionable. The technology is being used in retail to understand customer traffic patterns, demographics, and prevent shrinkage. And the same computing power is also being used to make cities smarter and safer through NVIDIA Metropolis.
Using Commercial Video as a Sensor to Gather Actionable Data
Often, a piece of technology serves only one purpose for a retailer. Security cameras serve as theft deterrents that capture and save video, but gathering data and making actionable inferences from it are outside of their capabilities. NVIDIA technology can make the actual video recordings seem obsolete because of the amount of data that can be collected and inferred from the pictures in real time.
“It essentially becomes a powerful video sensor that mounts on a wall, but no video leaves it, only data,” Adam said. “That’s a good example of where we see the market going, almost ignoring the fact that it’s video and focusing on the valuable data we can gather from it.”
With NVIDIA’s Jetson TX1 embedded into a security camera, it becomes a powerful AI machine capable of not only carrying out complex tasks but also learning over time what data is most useful.
“There are approximately 500 million security cameras in the world, and that number is expected to double by 2020,” Adam said. “That’s one billion cameras streaming video 24/7, where meaningful insights can be extracted to make cities smarter and safer. By applying deep learning, these cameras become intuitive sensors that can judge activity, demographics, or dwell time.”
One NVIDIA technology partner — MotionLoft — makes sensors that leverage video capabilities and uses the Jetson TX1 to track both in-store and property traffic numbers, giving retailers a complete picture of what is going on in and around their store. From heat maps and customer path data inside a store location to bike and vehicle traffic outside of it, MotionLoft’s video sensors are a shining example of what NVIDIA technology can do.
“Our products can see like the human eye, and they leverage the neural networks and machine learning to look for different types of objects,” said Chris Garrison Vice President of Business Development at MotionLoft. “Once it identifies an object, whether it is a person or a car, the sensor tracks it until it leaves, processes all of that information inside the device, and then sends the metadata back to the cloud.”
That data can unearth actionable information, like whether outside traffic patterns show an opportunity to optimize store hours to capture more walk-in traffic, or if a sales agent needs to be placed in a point of congestion that has been identified on a heat map.
GPU Goes Beyond Graphics to Accelerate Cloud Computing & Insights
Supercomputer processors and neural networks that are built into the cameras or on-premises appliances can analyze and recognize insights without having to push massive amounts of data to the cloud to be sorted through later. That ability can make a retailer’s cloud become exponentially more efficient because the analyzed data can be directly uploaded from the camera to the cloud, so it is not bogged down by unnecessary information.
“The edge of our technology is the camera itself, and we are building a true edge-to-cloud computing architecture,” Adam said. “We work with companies that are deploying rich vision solutions, but they don’t have the network to stream the video to a cloud. Our technology allows them to compute the AI right inside the camera.”
And if that data is pushed to the NVIDIA GPU Cloud (NGC), a business can take advantage of the vast amount of computing power to process even more information. With NGC, data scientists can create large neural networks that can provide insights much faster than traditional CPU clouds.
And GPU cloud computing is available on many major hosting platforms, including AWS, Microsoft Azure, Google CloudPlatform, and IBM Cloud.
Solving Loss Prevention Problems Through Deep Learning
Another major problem that NVIDIA technology is helping solve with its AI products is shrinkage. Using NVIDIA’s processors, companies like ThirdEye Labs are innovating ways to not only keep shelves stocked, but also detect and report product theft in real time.
Based on video camera streams, this automated technology is currently testing behavior and object recognition technologies to monitor for suspicious activity that can be identified as it is happening. And it is also helping to detect theft at the point of sale, analyzing checkouts to detect fraudulent scans and scan avoidance, two types of theft that can be incredibly difficult to identify in real time.
ThirdEye Labs is using the same technology to monitor store shelves to prevent stock from becoming depleted. The cameras can detect stockouts or other anomalies, such as products being put in the wrong place, and alert associates immediately.
Other companies are using NVIDIA processing power to detect real-time security threats, like Deep Science AI, which has developed an AI Surveillance (AIS) platform that can spot the presence of a mask or a gun being used in a robbery and alert authorities immediately.
And with those kinds of safety and security applications, it is no wonder that NVIDIA’s sights are set on an even bigger space than a retail environment.
Focusing on Businesses to Help Make Entire Cities Smarter
The NVIDIA Metropolis platform is the culmination of all of the company’s technologies and is built to be the foundation of the AI City. Through existing camera infrastructure, images can be monitored and analyzed to find efficiencies in areas like parking, traffic flow, or even law enforcement.
“I could say, ‘Show me bicyclists who have taken a right at this intersection in the last 10 hours, and the technology could process all of that video and converge it, showing you all of that information.” — Adam Scraba, Global Business Development Lead at NVIDIA
“When we think about an entire urban area, there are big problems to tackle,” Adam said, “Two things that we can impact are the efficiency of systems and — just as important — security. And we go well beyond classic security and surveillance as we work with law enforcement on safety.”
The technology is so smart and responsive that videos can be queried that match a particular set of data input into the system and will produce results quickly.
“I could say, ‘Show me bicyclists who have taken a right at this intersection in the last 10 hours,’ ” Adam told us. “And the technology could process all of that video and converge it, showing you all of that information.”
And Metropolis is a developer platform that works with an ecosystem of partners and software vendors to create productive solutions based on deep learning that wasn’t available before.
Positioned as an Innovator in Edge-to-Cloud AI Solutions
NVIDIA is still incredibly popular with gamers, making GPUs that are second to none to help players navigate fantasy worlds with incredible clarity. But the company has managed to successfully transfer the power it took to make those virtual worlds work into the real world to make cities run, and retailers analyze data to grow.
“With our technology, the path to ROI for retailers can be so much shorter,” Adam said. “Some of these technologies can be used to create immediate value. When you can analyze data quickly, you can do so many things and answer so many interesting questions.”
Smart and safe cities need AI as their canvas. And using NVIDIA technology to gather insights is like painting the Mona Lisa in a fraction of a second.