In the era of IoT, the true value of the Internet of Things lies in the increasing connections among everything. The explosion of massive data requires the use of extensive machine learning and artificial intelligence for analysis.

The rapid development of edge computing has accelerated the migration of computing power to the edge, and AI is gradually expanding from central nodes to the edge, closer to data sources and business sites.

When edge intelligence combines with edge computing and artificial intelligence, it can effectively perform real-time processing of small data, carry out AI model inference, and send results back to the cloud. This “cloud-edge-end” collaborative edge intelligence architecture addresses issues such as massive data processing, real-time response, and data security in current AI applications, laying the foundation for AI applications in more industries.

According to Deloitte’s analysis, current AI computing has been applied in various scenarios such as manufacturing, government, retail, telecommunications, and healthcare. Clearly, edge intelligence plays an important role in expanding the boundaries of AI, significantly enhancing AI’s adaptability to diverse business scenarios on-site, thereby better supporting business operations and creating more value for customers.

01

Software Empowerment Accelerates AI Application Development

If one wants to design an energy-efficient edge AI system without the support of appropriate tools and software provided by embedded processor vendors, while also accelerating time to market, this task is bound to be lengthy and tedious.

At the same time, enterprises will face a series of challenges, including selecting the appropriate deep learning model, training and optimizing the model against performance and accuracy targets, and learning to use dedicated tools for deploying models on embedded edge processors.

At the hardware level, NVIDIA has proposed the Jetson product family, creating a leading edge AI platform. Meanwhile, at the software level, NVIDIA has also launched the NVIDIA TAO Toolkit and Metropolis SDK to help enterprises accelerate the creation, deployment, and expansion of AI and IoT applications from edge to cloud.

Among them, the TAO Toolkit is NVIDIA’s AI model adaptation platform, which simplifies and accelerates the creation of artificial intelligence. The TAO platform brings together various NVIDIA technologies needed to accelerate the model adaptation process, including the Transfer Learning Toolkit (TLT). Models trained using the TAO toolkit can be deployed on any NVIDIA platform, including Jetson.

Based on NVIDIA’s TAO software stack, ThunderSoft provides an AI full-process application platform called Model Farm, which includes data collection, data cleaning and labeling, training set organization, training implementation, inference model deployment, and secondary verification of inference results. Through the AIOPS approach, it achieves a closed loop of data and business scenarios, helping customers quickly build scenario-based AI industry solutions.

Xu Chunliang, Senior Product and Solutions Director at ThunderSoft, gave a thematic presentation on “AI Full-Process Application System Based on TAO,” detailing how to achieve rapid development of AI applications based on Model Farm.

NVIDIA Metropolis SDK provides a set of end-to-end application development tools and frameworks, enabling efficient application development and accelerated deployment from data centers to edge. Combined with NVIDIA’s EGX enterprise platform, it better achieves cloud-edge collaboration and ensures stable and reliable enterprise-level operations.

Cui Xiaonan, NVIDIA Developer Development Manager, conducted online training on “Using Metropolis SDK to Quickly Develop and Deploy AI Applications,” helping you learn more about Metropolis SDK development techniques.

02

Scenario-First Accelerates Edge AI Implementation

In the construction of smart cities, AI applications have permeated various aspects such as urban management, transportation, and people’s livelihoods. As smart cities evolve towards refinement and community-based approaches, the application of edge intelligence in scenarios such as smart transportation, emergency response, and urban safety is increasing. IDC predicts that in the coming years, edge intelligence spending in urban operation and management scenarios will maintain a growth rate of 25%.

For example, in smart transportation, the Maricopa County Department of Transportation (MCDOT) in Arizona manages the 14th busiest traffic area in the United States, facing serious traffic congestion issues. NVIDIA Metropolis partner NoTraffic provides the necessary answers through an AI-based traffic management platform that optimizes processes based on real-time utilization and demand.

AI sensor units are installed at each intersection, providing road user detection and classification through the fusion of machine vision and radar. These sensors utilize NVIDIA Jetson edge AI devices and the NVIDIA Metropolis AI framework for video processing at intersections. NoTraffic’s roadside units can detect and classify road users, including cars, buses, trucks, bicycles, pedestrians, and even emergency vehicles under any lighting or weather conditions.

Using edge computing to optimize processed data can improve traffic signal lights, save bandwidth, and reduce latency. Urban engineers can use customized dashboards for data viewing and real-time analysis, enabling functions such as collision prediction and defining specific road user priorities.

Compared with common edge computing products, NVIDIA Jetson embedded GPU product line has advantages of high computing power, low power consumption, and low latency, combined with its cloud-edge integrated software stack and toolchain, allowing models trained in the cloud to be easily deployed to the edge and achieve dynamic online maintenance and updates.

As an NVIDIA Jetson Elite Partner, Tianjun Technology has developed a complete end-to-end solution based on Jetson for smart transportation vehicle-road collaboration scenarios, including edge computing MEC, intelligent perception cameras, multi-sensor temporal and spatial synchronization, and fusion perception solutions. This solution is widely used in holographic intersections, smart highways, autonomous parking, etc., achieving industry-leading precision and accuracy levels while maintaining the industry’s lowest system latency.

Liu Junchuan, General Manager of the Robotics Division at Tianjun Technology, will introduce you to the end-to-end smart transportation vehicle-road collaboration solution based on Jetson and Deepstream. The content is exciting and not to be missed.

Conclusion

As network architectures and tools continue to adapt and be compatible with embedded systems, more and more artificial intelligence applications can run directly on edge devices, making “edge AI” a hot topic in the industry.

Looking ahead, the rapid growth of edge AI, the integration of AI and edge computing, and accelerated implementation will drive us into a new era of AI in the “Internet of Everything”.