EyeCloudAI CDK is an open source edge-AI camera development kit for applications where privacy, real-time performance, reliability, and flexibility are paramount. With this camera-development kit (CDK), developing and deploying custom edge-AI camera solutions is easy. We’ve taken care of the AI optimization, sensor tuning, optics, tooling, and supply chain management so you can focus on your application.
EyeCloudAI CDK is based on the Intel Movidius Myriad X VPU and is compatible with the Intel Distribution of OpenVINO Toolkit. We are offering two versions of our core board. The first model is designed to be used while connected to an external host device or integrated into a larger system, be it a laptop, a single-board computer (SBC), a robot, or something else. The second model supports independent, stand-alone operation by allowing the CDK to boot from an 8 GB eMMC expansion board and access AI models stored there.
You can pair either version of the core board with any of three image signal processor (ISP) sensor modules: a 2 MP rolling-shutter module for normal applications, an 8 MP rolling shutter module for high-resolution applications, or a 2 MP global-shutter module (based on the ON Semi AR0234) for high-speed applications.
Each camera module is an independent, 38 x 38 mm board that connects to the EyeCloudAI CDK core board by means of an FFC cable. This design provides maximum flexibility while reducing the amount of work required to integrate these sensors into your final product.
The Intel Distribution of OpenVINO Toolkit allows you to take advantage of most pre-trained models from the OpenVINO Model Zoo or to create your own models, which you can then load onto EyeCloudAI CDK with our SDK or our easy-to-use Windows- and Linux-compatible graphical software. If you are working with the version that includes an 8 GB eMMC expansion board, you can update both the firmware and the set of available AI models using any UVC-compatible tool.
In addition, EyeCloudAI CDK allows you to deploy two convolutional neural network (CNN) models to a single core board and configure them to operate in series. By using the output of the first model as input to the second, EyeCloudAI CDK can produce complex results independently, without having to move data to and from a computer. In the example shown below, a license-plate recognition camera passes the output of a plate-detection model to a license-recognition model automatically.
|VPU||Intel Movidius Myriad X MA2085|
|Vector Processors||16 SHAVEs|
|Neural Network Capability||Two neural compute engines (up to 4 TOPS)|
|RAM||8 Gbits, LPDDR4 (1600 MHz, 32-bit)|
|Power Supply||5 V / 2 A|
|Input Data Interface||MIPI 4 lane|
|Output Data Interface||USB Type-C 3.0 & 2.0 selectable by cable orientation|
|Expansion Interfaces||GPIO, I²C, SPI, SD/SDIO, UART, 2x two-lane DSI (up to 2.5 Gbps per lane)|
|Dimensions||38 mm x 38 mm|
|Cable||30 cm USB Type-C|
|2MP RS Sensor Module||2MP GS Sensor Module||8MP RS Sensor Module|
|Sensor||SmartSens SC200AI||ON Semi AR0234CS||SmartSens SC8238H|
|Frame Rate||30 FPS||120 FPS at 1 MP or 60 FPS at 2 MP||30 FPS|
|Resolution||1920 x 1080 (2K)||1920 × 1080 (2K)||3872 × 2180 (4K)|
|Shutter||Rolling shutter||Global shutter||Rolling shutter|
|Field of View||Diagonal 60°||Diagonal 63°||Diagonal 58°|
|Output Data Interface||MIPI 2 lane||MIPI 4 lane||MIPI 4 lane|
|Dimensions||38 x 38 mm||38 x 38 mm||38 x 38 mm|
|Weight||12 g||12 g||12 g|
|Cable||4 cm FFC||4 cm FFC||4 cm FFC|
|EyeCloudAI CDK||OAK-1||Intel NCS2|
|Camera Sensor(s)||Semi AR0234CS, SmartSens SC200AI, or SC8238H||IMX378||No|
|NPU Performance||4 TOPS||4 TOPS||4 TOPS|
|Neural Compute Engine||2 x||2 x||2 x|
|Output Data Interface||USB Type-C 3.0 & 2.0 selectable by cable orientation||USB 3||USB 3.0 Type-A|
|Reserved Interfaces for Customization||GPIO, I²C, SPI, SD/SDIO, UART, 2x two-lane DSI (up to 2.5 Gbps per lane)||No||No|
|Camera Sensor Resolution||Selectable from 2 MP (1920 x 1080) to 8 MP (3872 x 2180)||Fixed 12 MP (4056x3040)||No|
|Hardware Video Encoding||H.264/H.265 at 2MP (30 PFS) & MJPEG at 2MP (30 FPS)||4K H.265 encoding at 30 FPS||No|
|Max Framerate||120 FPS||60 FPS||No|
|User-Friendly Software to Update Firmware & AI Models||Yes||No||No|
|eMMC||Via expansion board||No||No|
|Bootable From ROM||Via expansion board||No||No|
|Capable of Running Two CNN Models in Series||Yes||No||No|
|Weight||9 - 12 g per board||56.7 g||77.8 g|
|Size||38 x 38 mm||45 x 30 mm||27 x 72.5 mm|
|Open Source||Schematic + software||Hardware + software||No|
|Price||Starting at $148 with sensor module||Starting at $149||$79|
EyeCloudAI CDK works with Intel’s distribution of OpenVINO Toolkit and is compatible with frameworks such as ONNX, TensorFlow, Caffee, MXNet, Kaldi, and PaddlePaddle. Our SDK is compatible with Raspberry Pi flavors of Linux for ARMv7 and ARMv8.
You can find the EyeCloudAI CDK documentation, software, and hardware design files in our GitHub repository. If you have a question, please reach out using the Ask a technical question link on our project page.
Produced by EyeCloud in San Jose, CA.
Sold and shipped by Crowd Supply.
A variant of the EyeCloudAI CDK core board designed for independent, stand-alone operation. Includes an 8 GB eMMC expansion board that allows it to boot from local storage. Measures 38 x 38 mm and comes with a 30 cm USB Type-C cable.
An independent 2 MP rolling-shutter sensor with a changeable M 12 lens. If your applications do not require high-speed capture or high-resolution imagery, this is a good, cost-efficient option. Requires an EyeCloudAI CDK Core Board. Measures 38 x 38 mm.
An independent 2 MP global-shutter sensor, based on the Onsemi AR0234CS, that captures up to 120 frames per second through a changeable M 12 Lens. Recommended for applications that involve production lines, transportation systems, and other scenarios that require the ability to capture fast-moving objects. Requires an EyeCloudAI CDK Core Board. Measures 38 x 38 mm.
An 8 MP rolling-shutter sensor with a changeable M 12 Lens. Recommended for applications that require high-resolution imagery. Requires an EyeCloudAI CDK Core Board. Measures 38 x 38 mm.