Partnership yields AI-based 3D sensing reference designs for AIoT devices

Claiming industry-first approaches for biometric access control, 3D electronic locks, and intelligent sensing across vertical markets, Ambarella, Inc., Lumentum, and ON Semiconductor have developed two joint AI-based 3D sensing reference designs that are designed to accelerate AIoT device deployment. These reference designs – Saturn for electronic locks and Vision+ for biometric access control readers – build on the companies’ previous joint solution for contactless access systems.

The reference designs leverage Lumentum’s high-performance VCSEL array illuminators and an ON Semiconductor image sensor using Ambarella’s AI SoC. The result is higher levels of accuracy and more intelligent decision-making for biometric access control, 3D electronic locks and other intelligent sensing applications. In addition, the high level of integration significantly lowers system power consumption and thermal design requirements as well as enables much smaller product form factors, said Ambarella.

In addition to the targeted biometric access control and electronic lock applications, these new reference designs can be used in applications found in smart cities, smart buildings, smart homes, and intelligent healthcare.

Image: Ambarella

The new reference designs are a natural progression from the previous collaboration, said William Xu, director of marketing, Ambarella. “Our first joint reference design, called Janus, was introduced in 2020. It was the industry’s first single-camera structured-light solution for biometric access control readers, as well as time-attendance and mobile-payment applications.”

Xu said subsequent engagements with OEMs led to the development of its first new joint reference design, Saturn, which is for hospitality locks and residential applications.

The companies claim the Saturn reference design is the AIoT industry’s first to integrate AI processing for both single-camera structured-light sensing and fast-boot video processing. Based on the Ambarella CV25 CVflow AI vision processor, it includes a structured-light camera powered by Lumentum’s VCSEL technology and the AR0237CS 2 MP RGB-IR image sensor from ON Semiconductor.

“At the same time, it became apparent that OEMs needed a complete product portfolio, for applications/deployments that require higher resolution, including for anti-spoofing and longer range,” he said. “The three companies again collaborated together for our second new joint reference design, called Vision+. For this reference design, ON Semiconductor built a new 4K RGB-IR image sensor, Lumentum upgraded its VCSEL illumination solution to support longer distances for 3D sensing, and we used Ambarella’s CV22 AI vision SoC to support this reference design’s image-signal and AI processing requirements.”

The Vision+ reference design targets biometric access control readers. Claimed as the AIoT industry’s first 4K solution to perform single-camera, 940-nm structured-light sensing based on RGB-IR technology, it is also said to be the first to leverage a single-chip solution for depth processing, AI processing, and video processing. Based on the Ambarella CV22 CVflow AI vision processor, it includes single-camera structured-light sensing powered by Lumentum’s VCSEL technology, featuring a range of two meters, and a 4K (8MP) RGB-IR CMOS image sensor from ON Semiconductor.

In addition to the two reference designs, 3D sensing development kits are available, addressing different application-specific requirements. For intelligent sensing applications, Ambarella’s CV2 series CVflow AI vision processor-based development kits can be supplied with ToF sensor adapters that are powered by Lumentum’s VCSEL arrays. An optional 4K RGB-IR image sensor adapter is also available.

Xu said the combination of the joint reference designs and development kits now enables OEMs to build a complete product roadmap of smart building applications. These include sliding door sensors, biometric access control readers, anti-tailgating devices for enhanced security, and occupancy sensing cameras that provide real-time occupancy intelligence for decision making.

Click for larger image. (Image: Ambarella)

In addition, the open software development kit (SDK) for Ambarella’s CVflow AI vision processors enables the integration of third-party applications. It also allows OEMs to address differing regional requirements with a single platform (for example, using different AI algorithms in different regions).

The Linux-based SDK offers a range of features, including support for different 3D modalities such as structured light and ToF, built-in Arm processors, an integrated ISP to handle challenging scenes, including high dynamic range and low light, and high AI inference capabilities on the camera device to enable multiple neural networks running simultaneously. Integrated security includes secure boot, OTP to store encryption keys, and Arm TrustZone technology for secure encryption key comparisons.

The SDK also offers custom application development capabilities and AI tools to support deep neural network  (DNN) development and migration. The joint reference designs and development kits are expected to be available in June from Ambarella.