Metropolis: Intelligent Deep Learning-Based Edge-to-Cloud Video Platform

NVIDIA announced that Verizon has joined over 100 companies using NVIDIA Metropolis to create safer, smarter, and more sustainable cities. NVIDIA Metropolis is NVIDIA's edge-to-cloud video platform that enables the development of intelligent, fast deep learning-based applications.

Verizon is a leading technology company providing the most reliable network services in the United States. Verizon's Smart Communities group is collaborating with multiple cities to connect communities and develop them for the future. A concrete example includes intelligent cameras based on NVIDIA Jetson mounted at street lights and other strategic locations throughout cities for monitoring urban conditions.

David Tucker, who oversees product management at Verizon's Smart Communities group, stated, "LED street lights provide significant cost savings in terms of operational expenses and are being rapidly adopted globally," adding, "Cities will expand their lighting infrastructure to establish smart city platforms, and through this, they will integrate various current and future applications to enhance efficiency and develop diverse citizen services."

Verizon designates these mounted intelligent cameras as video nodes. These cameras leverage Jetson's deep learning capabilities to analyze multiple video data streams, improving traffic flow, enhancing pedestrian safety, and resolving parking issues in urban areas, among other functions.

Beta testing using proprietary datasets and models generated from neural network training is expected to conclude in the eastern and western coastal regions of the United States, and Verizon plans to announce detailed information regarding full-scale commercialization in the near future.

The NVIDIA Metropolis platform, which was unveiled last year, includes various tools, technologies, and support for building deep learning applications across nearly all sectors, from traffic and parking management to public safety and urban services.

High-performance deep learning inference tasks are performed at the edge where NVIDIA Jetson embedded computing platforms are deployed, as well as through servers and data centers equipped with NVIDIA Tesla GPU accelerators.

Verizon's video nodes utilize Jetson TX1 to collect and analyze data from the farthest edge of the city network. The supercomputer located on the module accelerates deep learning at the edge, enabling real-time video analysis. Through this edge computing process, data analysis can be performed more efficiently and at near real-time speeds, while also reducing the high cost of streaming and storing video through LTE and Wi-Fi networks.

The video nodes capture and classify objects such as vehicles, bicycles, and pedestrians, and identify their interactions at near real-time speeds, providing city officials with 24-hour data streams covering nearly all information ranging from illegal right turns at red lights to pedestrian movements outside crosswalks and parking lot conditions.

David Tucker stated, "Through Jetson, we discovered that we can create a consistent deep learning perspective that leverages GPUs to span the entire stack from cloud to edge."

While Jetson-based nodes quickly detect the movement of fast-moving vehicles and bicycles and handle other real-time tasks at the edge, when this data reaches the cloud, it can also be utilized for predictive analytics.

David Tucker explained, "We are moving forward in a direction where we can capture what happens at intersection A and understand in real-time the impact it will have on intersections B and C several blocks away."