EyeCloudAI will soon go to Crowd Supply to launch an open-spec, Pi-compatible “CDK” edge AI camera kit based on Intel’s Myriad X VPU with 1GB RAM and a choice of 2MP rolling or global shutter or 8MP rolling shutter sensors.

EyeCloudAI, which successfully launched a series of OpenNCC embedded vision AI cameras on Kickstarter last fall, will soon launch a Crowd Supply campaign for a smaller, more modular EyeCloudAI CDK camera kit. Like the OpenNCC cameras, the EyeCloudAI CDK is built around an up to 4-TOPS Intel Movidius Myriad X VPU, which has appeared on numerous edge AI computers, robots, and camera kits such as Adlink’s Apollo Lake based NEON-1000-MDX.

EyeCloudAI CDK (left) and Core Board, adapter board, and three sensor board options
(click images to enlarge)

The EyeCloudAI CDK is designed for applications “where privacy, real-time performance, reliability, and flexibility are paramount,” says the company. “We’ve taken care of the AI optimization, sensor tuning, optics, tooling, and supply chain management so you can focus on your application.”

As with OpenNCC, the hardware is open source and there is an open SDK based on Intel’s OpenVINO toolkit that offers compatibility with frameworks such as ONNX, TensorFlow, Caffee, MXNet, Kaldi, and PaddlePaddle. Unlike OpenNCC, EyeCloudAI specifically claims compatibility with ARMv7 and ARMv8 based Raspberry Pi compatible Linux distributions.

No pricing has been set for the EyeCloudAI CDK, which we saw on Hackster.io, but the base-level model at least will likely be more affordable. You can choose from three different USB Type-C connected, ISP-ready models all using the same Myriad X equipped Core Board, but with a choice of different sensors. There are two 2-megapixel models, with rolling and global shutters, respectively, as well as an 8MP rolling shutter option .

By comparison, only the base-level, $399 ($299 early bird) OpenNCC series camera uses a USB interface. The others are GigE Ethernet cameras with PoE: a standard $499 model and a $1,495 to $1,795 model with an IR thermal sensor and an HDMI port. All three OpenNCC cameras have global shutters for up to 120fps imaging.

Internally, the EyeCloudAI CDK Core Board communicates with the separate sensor boards via MIPI-CSI. The 38 x 38mm, 9-gram Core Board has 1GB LPDDR4-1600 and a 5V/2A power supply. Expansion interfaces include GPIO, I2C, SPI, SD/SDIO, UART, and a pair of two-lane MIPI-DSI connections for up to 2.5 Gbps per lane. There is also an adapter board between the Core Board at the sensor board.

EyeCloudAI CDK sensor specs (left) and earlier OpenNCC
(click images to enlarge)

The 2MP rolling shutter sensor board uses a SmartSens SC200AI sensor and communicates via 2-lane MIPI-CSI. The other two options use 4-lane CSI. The 2MP global shutter model has an ON Semi AR0234 sensor while the 8MP rolling shutter model has a SmartSens SC8238H sensor. All three offer M12 lens support.



Car and plate detection using EyeCloudAI CDK

 
Further information

The EyeCloudAI CDK is “coming soon” to Crowd Supply where you can sign up to be notified. You can also check out the GitHub page and the EyeCloudAI website.