Chip designer Ambarella has announced a new computer vision chip for artificial intelligence at the edge of a computer network, such as in smart cars or security cameras.
The new CV28M camera system on chip (SoC), the latest in the CVflow family. It combines advanced image processing, high-resolution video encoding, and computer vision processing in a single, low-power chip.
Ambarella packed a lot of AI processing power in the chip because of the way that computer networks will evolve in the future as everything gets connected to the internet. Since the networks could get inundated with data traffic, many applications like self-driving cars will have do their processing at the edge of the network, or in the car itself, rather than interact a lot with datacenter processors.
That means that the sensors and image processors in edge devices will have to be very powerful, said Chris Day, vice president of marketing at Ambarella, in an interview with VentureBeat.
“We’re working on new markets, including IP security cameras, consumer home monitoring cameras, drones, automobiles, and autonomous vehicles,” Day said. “This one is at the low end and it enables us to do applications at different price points and features.”
He said the applications for the CV28M could include home security, retail monitoring, consumer robotics, and occupancy monitoring. As privacy laws become more strict and people care about their privacy more, it’s also important to do processing at the edge, rather than said private data to the cloud.
One example of where the chips could be used is cameras that monitor the elderly, detecting whether a person has fallen down in their home. The camera would not record anything while the person is fine. But if the sensors detect a fall, the camera could turn on and ask the person if they are OK, Day said.
Some elderly-monitoring devices use radio to detect a fall, but Day said the need to communicate with a person to eliminate false positives means that the application needs a video communication capability.
“The privacy of the person you’re monitoring is not impacted,” Day said. “You can’t and you don’t want to be monitoring them 24 hours a day. But you do want to know when there is a problem.”
For internet-connected security cameras, the CV28M features AI-based rate control to optimize image quality while reducing video storage and network bandwidth requirements.
Ambarella’s AI Timelapse scene-aware recording avoids the time needed to scan through video timelines to retrieve moments of interest.
“AI time lapse reduces the storage requirements and means you don’t have to record and watch a whole lot of video when nothing is happening,” Day said. “In retail monitoring, we can actually monitor shopper behavior, see how many people are in the store, and where people are congregating in a particular area. We can create heat maps based on that.”
In consumer robotics applications, the CV28M can be connected to sensors such as visible, structured light, and time-of-flight (ToF) to capture, and then process, the data required for navigation. With the pandemic, the cameras could be used to detect whether people are appropriately standing six feet apart or more, Day said.
The CV28M can also do efficient video encoding in both AVC and HEVC formats. The chips are fabricated in 10-nanometer manufacturing.
The CV28M chip shares a common SDK and computer vision (CV) tools with Ambarella’s CV25, CV22,
and CV2 CVflow SoC families. It has a dual-core 1-GHz Arm Cortex-A53 processor, and it only consumes around 500 milliwatts of power.
The Santa Clara, California-based company, which went public in 2012, started out as a maker of low-power chips for video cameras. But it parlayed that capability into computer vision expertise and launched its CVflow architecture in 2018 to create low-power artificial intelligence chips. Now it has 800 employees and is competing with the likes of Intel and Nvidia, though with a focus on low-power applications. The company generated $220 million in revenue in 2019.
The CV28M chips are available in sample quantities now. But the company did not say when they would be in mass production.
How startups are scaling communication:
The pandemic is making startups take a close look at ramping up their communication solutions. Learn how