We’ve previously written about Hailo-8 AI accelerator delivering up to 26 TOPS at low power (3 TOPS per watt), and found in AI edge systems such as Foxconn BOXiedge server powered by a Socionext SynQuacer SC2A11 24x Cortex-A53 cores SoC and capable of analyzing up to 20 streaming camera feeds in real-time for video analytics.
Leopard Imaging has now created a much more compact Edge AI processing solution with EdgeTuring based on Hailo-8 M.2 card and Socionext SC2000 4K camera SoC, and designed for low-power video analytics.
EdgeTuring features and specifications:
- Camera SoC / Image Processor – Socionext SC2000 quad-core Cortex-A7 processor @ 650MHz with 4Kp30 H.265/H.264 encoder, up to 1.2 Gigapixel per second processing, LPDDR-3 memory (PoP)
- AI Module – Hailo-8 M.2 Module for up to 26 TOPS
- Dual camera but only one sensor works for AI
- Sensors – 2x Sony IMX477 CMOS Image Sensor, diagonal 7.857 mm (Type 1/2.3)
- Resolution – 3840 (H) x 2160 (V)
- Pixel Size – 1.55 x 1.55μm
- Video Output: 4K @ 30 fps with EIS / 4K @ 15 fps with HDR
- Baseline: 40.70 mm
- Lens: FOV 43.4° horizontal (each video)
- Storage – 2x MicroSD card slots
- Video Output – Micro HDMI
- Networking – Ethernet RJ45 port
- USB – USB 3.0 Type-C port, micro USB 2.0 port, micro USB port for UART console
- Power Supply – 12V DC in via 2-pin connector
- Dimensions – 80.70 (L) x 38.00 (W) x 93.33 (D) mm
- Weight – 242 grams
- Temperature Range – Operating: 0℃ to +60℃; storage: – 20℃ to +70℃
The mini video analytics server runs Linux using integration with two Amazon AWS Kinesis services: Amazon Kinesis Video Streams (KVS) and Amazon Kinesis Data Streams (KDS). EdgeTuring is said to have an accuracy ranging from 95% to 99% for several deep learning-based computer vision applications, such as object detection, and image classification, and all that at a higher speed, lower power consumption and costs compared to other, unnamed, competing solutions.
It should be noted that SocioNext SC2000 camera SoC already supports AI acceleration on its own with hardware-assisted face detection and smile/blink detection. But the Hailo-8 M.2 card brings TensorFlow and ONNX compatibility, and a higher level of flexibility and performance enabling more features applicable to industrial automation, IoT, smart devices and smart retail use cases.
Leopard Imaging’s EdgeTuring is currently sold for $899 in single units. You can find more details and/or request to purchase a sample on the product page.