Potential to create AI-ready wide-angle footage with embedded contextual data.
Processor IP licensor CEVA has invested $10 million in Immervision, a Canada-based startup that develops hardware and software for wide-angle lens camera systems. Combining Immervision’s Data-In-Picture technology with sensor fusion capabilities from Hillcrest Labs, which CEVA acquired last month, could enable creation of AI-ready wide-angle footage with contextual metadata embedded in each video frame.
In an exclusive interview with EETimes, CEVA’s VP market intelligence, investor and public relations Richard Kingston explained that CEVA’s $10 million investment in Immervision will allow the company to continue to develop its optical hardware and software.
“Immervision has developed a new lens which is flatter than a fisheye lens, with the same field of view, but with much less distortion and skewing in the corners,” Kingston explained. “Part of that work is done by the lens itself, and then there is a lot of software that works alongside the lens in order to recognise objects and improve the image. That software works with any type of lens,” not just the Immervision lens.
Immervision’s algorithms can flatten wide-angle footage, removing distortion at the edges, and embed sensor fusion data in wide-angle video frames (Image: CEVA)
Fisheye lenses are widely used in surveillance systems and other applications as they have a very wide field of view, but they typically suffer from distortion around the edges of the image. Immervision’s algorithms flatten wide-angle images and remove distortion and skewing to create intelligible photos that preserve object proportions and straight lines. Also, the company’s Data-In-Picture technology allows metadata such as timestamps to be embedded in the corners of video frames from the wide-angle footage, which are otherwise unused. The company’s technology has shipped in more than 50 million devices to date.
Immervision Data-In-Picture technology embeds metadata in the corners of wide-angle video frames (Image: Immervision)
Under the terms of the deal, CEVA will receive exclusive licensing rights for Immervision’s software.
“CEVA will effectively take over the software part of [Immervision’s] business — the software they develop will become CEVA’s IP to license [exclusively],” Kingston said. “CEVA will use its extensive existing sales team of around 40 people worldwide, who specialise in licensing IP, and bring Immervision’s IP to the CEVA customer base.”
There is a revenue share scheme in place and Immervision will also receive royalties as part of the deal, Kingston said. Moving forwards, Immervision will continue to license the optical hardware system directly, but licenses for the complementary software will be sold by CEVA. The two companies will work together to serve mobile phone manufacturers, camera manufacturers and OEMs developing surveillance systems and automotive ADAS (advanced driver assistance systems).
An important secondary part of the deal will see Immervision optimize its image processing software for CEVA’s computer vision DSP IP cores.
“We have about 50 customers using our vision DSPs in their chips. Today, most Immervision customers run the software on GPUs within an SoC, because it’s already there, and it’s easy to program,” said Kingston. “But the DSP’s performance is better than a GPU for these types of algorithms. So as part of the deal, Immervision will optimise their software for our vision DSP and then we’ll be able to offer a much more power-efficient and performance-enhanced alternative to what some of Immervision’s customers are doing today.”
CEVA acquired Hillcrest Labs, based in Rockville, MD, last month from InterDigital for $11 million. Hillcrest develops inertial and environmental sensor modules and sensor fusion software for consumer and IoT devices.
There is a potential synergy between Hillcrest’s technology and Immervision’s: sensor fusion data could be embedded into wide-angle camera footage, frame by frame, using Immervision’s Data-In-Picture technology. This could be used to embed timing data for synchronisation of frames from multi-camera security systems, or to embed contextual data such as where the image was taken, to allow artificial intelligence systems to better understand the images.
Both of CEVA’s recent deals support the company’s strategy of moving up the value chain, since both Immervision and Hillcrest deal directly with OEMs rather than semiconductor vendors. IP from both companies can run on SoCs that OEMs have already designed in.
Hillcrest Labs, acquired by CEVA last month, develops sensor modules and sensor fusion software (Image: Hillcrest Labs)
“[Compared to licensing processor IP], there’s no two-year chip development cycle, you can just license the software and run it straightaway,” Kingston said. “As a result of working with OEMs, we can get a better royalty rate than we would if we were working with semiconductor industry. So this is another way for us to move up the food chain and start licensing to OEMs directly; it opens up a door that we haven’t had in the past, as we have mainly worked with semiconductor vendors up to this point.”
With SoC design becoming more complex and difficult, CEVA’s approach also includes offering more than just processor building blocks.
“It’s a lot more difficult than people think it is to start developing these systems in-house,” Kingston said. “If you’re a processor company and you’re used to just dealing with hardware, when you start dealing with software as well it can become quite burdensome… If CEVA can offer all the software and the algorithms and the tools and everything you need, then we’re able to help our customers in a way that wasn’t possible in the traditional IP licensing or processor industry that has been around for a long time.”