OpenCV and Luxonis have launched the successor to the popular OAK-D all-in-one computer vision platform, offering all the features — albeit with a few trade-offs — for considerably less money as the OAK-D-LITE.
"With the previous OAK-D Kickstarter we enabled thousands to harness the power of Spatial AI," claimed Luxonis chief architect and chief executive Brandon Gilles, the man behind the campaign. "But too many ended up in the situation where they wished they could use it to solve their problem — because it's a cheat mode for so many problems — but it was simply too expensive to use it."
The OAK-D-LITE — pleasingly pronounceable as "oak delight" — aims to resolve that, dropping the price from $149 during the discounted crowdfunding campaign to just $79, both representing a claimed 50% discount over the devices' target sales price.
OpenCV launched its campaign for the original OAK-1 and OAK-D boards last year, with rewards priced at $99 and $149 respectively. It proved a staggering success: The initial $20,000 funding goal was reached within 20 minutes, and the campaign closed having raised nearly $1.4 million from over 6,500 backers.
Hardware was scheduled to slip in December that year, a target OpenCV hit — an unusual thing indeed in the world of crowdfunding — before shipping an OAK-D-IoT variant in May 2021 and Power-over-Ethernet (PoE) variants in July.
The OAK-D-LITE, OpenCV says, offers "all the powers of the OAK-D" but at a cheaper price point — yet with hardware including a 4K-resolution RGB autofocus camera, a depth-sensing stereo pair with 300,000 measurement points at distances of up to 33 feet, and a total of 20 computer vision processing elements including 16 "mini-GPU" vector processors provided by an Intel Movidius Myriad X chip, a depth processor capable of 200 frame per second operation, and two dedicated AI processors.
With its combination depth-sensing and visible-light cameras plus hardware accelerators, the OAK-D family offers a range of on-board computer vision capabilities — which, in its key selling point, don't load the CPU or GPU on the host device. OpenCV has shown off edge detection, feature tracking, object tracking, background subtraction, edge filtering, object pose detection, semantic segmentation, pose estimation, gaze estimation, vehicle and pedestrian detection, and more — including on-device anonymisation for both photography and video recording.
The OAK-D-LITE somehow manages to be even smaller than its predecessor, too: OpenCV claims the pocketable system is 25% of the volume and 50 per cent of the weight of the original OAK-D — making it better-suited to embedded and weight-sensitive applications, such as drone platforms.
There are, naturally, some trade-offs. The original OAK-D offers up to 1280x800 resolution depth capture from its paired OmniVision OV9282 sensors; the OAK-D-Lite uses OV7251 sensors and limits depth resolution to 640x480 - the most popular resolution among user, OpenCV claims, and one of the ways it was able to save "substantial cost" on the project. The RGB camera is also downgraded, from a Sony IMX378 supporting up to 60 frames per second to the lower-end IMX214 and a 30 frames per second limit.
Fully compatible with the existing DepthAI OAK ecosystem, software for the OAK-D-LITE is provided for Windows, macOS, desktop Linux, and the Raspberry Pi and NVIDIA Jetson single-board computers. OpenCV also boasts "plug-and-play" compatibility with the Robot Operating System (ROS) and ROS2, making it easy to integrate into robotics projects.
"OAK-D-LITE makes it so you can proof-of-concept your idea in hours," Gilles explains, "not the months of fighting and integrating disparate libraries, dependencies, hardware pieces, and drivers that plagues the industry. (Cough: We're looking at you, NVIDIA.)"
The OAK-D-LITE campaign is now live on Kickstarter, with physical rewards starting at $79 for early bird backers. Delivery, meanwhile, is expected to take place in December 2021.