Ambarella Unveils CV28M SoC, Powering Intelligent Edge Cameras with AI-Driven Video Encoding

Ambarella, a leading provider of computer‑vision processors, has introduced a new SoC that expands the company’s CVflow portfolio and targets a broader range of smart‑device applications demanding flexible edge‑AI sensing and processing.

The CV28M, the latest addition to the CVflow family, integrates advanced image processing, high‑resolution video encoding, and computer‑vision acceleration into a single low‑power package. Two capabilities that stand out for intelligent edge sensing and bandwidth efficiency are AI‑based rate control—optimizing image quality while trimming storage and network bandwidth—and AI‑Timelapse, which analyses scenes to record only moments deemed valuable, eliminating the need to sift through hours of footage.

Jerome Gigot, director of marketing for Ambarella, said, "AI-based bitrate control is the ability to use the Ambarella CVFlow AI engine to analyze every frame before encoding to find the most important regions of each frame (such as a face or a car for example) and be able to encode that specific region of the video with higher quality compared to other regions that don’t contain any important information, such as the background or other unimportant objects."

In addition, Chris Day, vice president of marketing and business development at Ambarella, explained to embedded.com that the new device uses the same AI engine and capability as its CV25, but is addressing new cost‑conscious markets and enabling a new class of smart edge devices for applications including smart home security, retail monitoring, consumer robotics, and occupancy monitoring.

"All around us, devices are becoming smarter, and with our newest CV28M SoC, our customers can develop a new generation of intelligent sensing cameras for a variety of new applications. In privacy‑sensitive applications—such as monitoring retail stores, workplaces, rental properties, or the elderly at home—edge‑based AI processing can support intelligent monitoring and fast decision‑making without the requirement to record or stream video to the cloud."

He said the main difference with the new camera SoC is that it uses a dual core Arm Cortex‑A53 as opposed to a quad‑core device. This results in greater efficiency for edge sensing AI, and up to 30% reduction in cost. He added that for many of the more consumer‑oriented embedded vision applications, they don’t need the full AI capability of the CV25, so the new CV28M provides a more targeted approach.

For AI sensing applications such as retail monitoring or occupancy monitoring, CV28M provides the AI performance to make all decisions in the camera, preserving privacy and avoiding heavy video processing running on back‑end servers. In consumer robotics applications, the CV28M can be connected to a wide range of sensors such as visible, structured light, and time‑of‑flight (ToF) to capture, and then process, the data required for navigation.

The CV28M delivers efficient video encoding in both AVC and HEVC formats. A high‑performance image signal processor (ISP) delivers outstanding imaging in low light conditions, and high‑dynamic‑range (HDR) processing extracts maximum image detail in high‑contrast scenes. The camera SoC includes a full suite of advanced cybersecurity features to protect against hacking, including secure boot, TrustZone, and I/O virtualization. Fabricated in 10 nm ultra‑low‑power process technology, the CV28M chip is optimized for wire‑free camera applications that require long battery life and small form factors.

The CV28M chip shares a common SDK and computer‑vision (CV) tools with Ambarella’s CV25, CV22, and CV2 CVflow SoC families, simplifying development of cameras with multiple price and performance options. A complete set of CV tools helps customers port their own neural networks onto CV28M and includes a compiler, debugger, and support for industry‑standard machine learning frameworks such as Caffe and TensorFlow, with extensive guidelines for convolutional neural network (CNN) performance optimizations.