Bench Talk

Join journalist Tommy Cummings for a weekly look at all things interesting, new, and noteworthy for design engineers.

Vision Artificial Intelligence (Vision AI) is a field of computer science that trains computers to replicate the human vision system.

Designers develop devices, such as face detectors, QR code scanners, etc., to identify and process objects in images and videos in the way that humans do.

It's not exactly a new technology. Computer vision can trace its roots back to the 1950s, when neural networks (NN) were designed for pattern recognition. In the 1970s, optical character recognition (OCR) was able to interpret handwritten text for the vision impaired. Since 2010, deep learning has helped computers train themselves and self-improve over time.

Today, computer vision has expanded with designers using the technology for automotive, healthcare, retail, smartphones, and more. Security and surveillance are also a big part of the technology, with AI automating the data gathering and security assessments.

In this week's New Tech Tuesday, we'll look at microprocessors from Renesas Electronics and kits from Basler and Xilinx for Vision AI development.

Renesas Electronics RZ/V2L High Precision Entry-Level AI MPUs are designed for basic AI-enabled applications, including home appliances, smart doorbells, surveillance cameras, point-of-sale (POS) terminals, and robot vacuum cleaners. The MPUs incorporate Renesas' AI accelerator, DRP-AI (Dynamically Reconfigurable Processor), to make embedded AI design easier and more power-efficient. The good news for designers: The RZ/V2L is package- and pin-compatible with the existing RZ/G2L general-purpose MPUs, allowing RZ/G2L users to easily upgrade to the RZ/V2L for additional AI functions without needing to modify the system configuration. Renesas also offers a complimentary DRP-AI Translator, a tool that automatically converts AI models into an executable format. Developers can leverage the DRP-AI while using familiar tools that allow them to immediately start using the RZ/V2L to evaluate AI design based on learning data.

The Basler AI Vision Solution Kit has full integration with cloud connectivity for rapid prototyping of AI-based Internet of Things applications. The kit can test the applications on an optimized vision system and access cloud services. Connecting to the cloud makes it easier to load or train neural networks. The kit also significantly reduces the complexity and need for expert knowledge of embedded hardware and software technology. The kit comes with a dart BCON for the MIPI camera module for the NVIDIA^® Jetson Nano^™ System-on-Module (SoM). This solution kit with cloud connectivity is a development kit for integrating a dart camera with a proprietary BCON for the MIPI interface. The AI vision kit contains a dart Mobile Industry Processor Interface (MIPI) camera module, an NVIDIA Jetson Nano development board, lens, and cable. The kit also includes all necessary drivers and software for cloud support that offers a complete plug-and-play design-in package.

Designers will find the Xilinx Kria^™ KV260 Vision AI Starter Kit an ideal out-of-the-box platform for vision application development without requiring complex hardware design knowledge. Developers of all levels can get applications up and running in under an hour with no field-programmable gate array (FPGA) experience needed. At that point, customization and differentiation can be added via preferred design environments, at any level of abstraction, from application software to AI model to FPGA design.

The carrier card allows various interfacing options and includes a power solution and network connectors for the camera, display, and microSD card. Target applications for the kit include smart city and machine vision, security cameras, retail analytics, and other industrial applications.

Like any artificial intelligence development, the technology provides multiple possibilities for consumers and businesses. Already much of AI Vision is being deployed in autonomous vehicles, cashier-less checkouts, medical diagnosis, and image labeling. The technology is still in development and learning modes, needing data to make humanlike determinations.