Vision AI is one of the fastest growing fields of embedded artificial intelligence, combining AI-enhanced voice tools and real-time analytics as a means to rapidly collect, process, and train large amounts of data. Masu. According to ITR Economics' forecasts, the vision AI market is expected to grow from $500 million in 2020 to $1.3 billion in 2025, with a compound annual growth rate of 22%.
The demand for embedded vision AI solutions is driven by the industry's continued movement away from over-reliance on cloud-connected communications in favor of AI solutions at the network edge. Edge-based AI systems enable end users to make informed decisions at a scale and speed previously unimaginable, while optimizing processing speed, energy consumption, and customer ease of use. must be provided.
As a leader in embedded technology solutions and committed to bringing the benefits of AI to a wide range of customers, Renesas recently announced its latest vision-based AI solution, the RZ/V2H microprocessor (MPU) platform. Developed for industrial, home, office, and smart city applications, this new approach to robotics automation allows designers to move to the edge and beyond without the cost, latency, and power penalties of cloud-based solutions. Quickly and easily integrate vision sensing systems into endpoints.
Evolution of vision AI processing performance
The new Renesas quad-core RZ/V2H MPU single platform accelerates multi-image processing by providing support for up to 4 cameras (up to 6), making it ideal for automated factory equipment, robotic control, transportation systems, and other end-users. Improve the accuracy of your application. Use the included USB port.
From a raw performance standpoint, the RZ/V2H platform includes Renesas' third-generation dynamically reconfigurable processors (DRPs). The unique DRP-AI3 AI accelerator delivers a significant 10x performance increase compared to previous models, enabling the new MPU platform to reach 80 trillion operations per second (TOPS). generation MPU (0.5 to 1.0 TOPS). .
Renesas has also developed OpenCV Accelerator, which leverages proprietary DRP technology to accelerate processing of OpenCV, an open source industry standard library for computer vision processing. The combination of DRP-AI3 and OpenCV accelerators powers both AI computing and image processing algorithms by processing data 16x faster than traditional CPUs.
Power-efficient design eliminates the need for fans and heat sinks in AI vision systems
Thanks to the advanced design of the DRP-AI3 accelerator, the new MPU platform increases power efficiency to 10 TOPS/W, resulting in 10x energy savings compared to previous solutions. This highly power-efficient design eliminates the need for fans and heat sinks required by competing solutions, significantly saving space, cost, and design time for AI applications running at the power-sensitive network edge.
Renesas achieved this breakthrough using a new approach to hardware and software, including coordination between AI accelerators and the main processor to quickly process various algorithms. Other power-saving innovations in the DRP-AI3 accelerator include making AI models lighter, including quantization to lower the bit weights of neural network data, and setting weight information to improve the efficiency of machine learning models. This includes pruning, which is a technique for skipping.
Renesas' Vision AI MPU platform makes customers' lives easier
To facilitate customer ease of use, Renesas has released an RZ/V2H evaluation board in addition to an AI application library of pre-trained models and an AI SDK. Together, these new tools will allow engineers to easily evaluate applications early in the design process without having to have extensive knowledge of AI. This includes preparing over 50 example applications that are free to download and use across multiple end uses. With 50 more application examples to be released soon, designers can take advantage of a wide range of potential use cases, including:
- Defect inspection: Monitor factory production to visually detect product defects
- Touchless industrial control: Replace physical controls with hand gestures
- Crop defense: Alert farmers before stray and wild animals damage crops
- Use of elevator: Enables touchless control and passenger counting
- Parking reservation: Track parking availability in real time
- Smart POS: Optimized retail checkout
In the future, we expect vision AI to be complemented by generative AI at the edge, bringing new levels of design complexity depending on specific data execution needs and desired performance levels. Today, generative AI remains a costly and power-intensive option, primarily used to process large datasets, but over time the two are working together to create a flexible, scalable, and cost-effective option. We believe that it promotes high decision-making.
Combining the two enables more complex image processing and also enables the integration of embedded vision systems with other AI processing models. Either way, the trend of AI reaching the network edge is inevitable.
