Edge AI camera runs Linux on quad -A53 SoC with Google Edge TPU

Edge AI camera runs Linux on quad -A53 SoC with Google Edge TPU

May 4, 2021 — by Eric Brown

— 81 views

Imago’s 5MP “VisionAI” camera runs Linux on a quad -A53 SoC accompanied by a Google Edge TPU for TensorFlow Lite and AutoML Vision Edge. Other features include 2GB DDR4, microSD, GbE, and DIO.

Imago Technologies GmbH announced a “freely programmable,” 5-megapixel edge AI camera designed for AI/ML and deep learning enabled image processing applications including pattern recognition, classification, anomaly or defect detection, and code reading. The VisionAI embedded camera runs Debian Linux on an unnamed quad-core, Cortex-A53 SoC clocked to 1.8GHz. Our guess is the i.MX8M Mini, but the same profile applies to a few other SoCs such as the Actions S900.

The SoC is paired with Google’s Coral Edge TPU AI accelerator. It is unclear if Imago is deploying the Edge TPU via the solderable, LGA form-factor Coral Accelerator Module or one of the M.2 or mini-PCIe modules. The Edge TPU offers 4-TOPS AI processing power using 0.5 watts for each TOPS (2 TOPS per watt).

VisionAI (left) and block diagram
(click images to enlarge)

The Edge TPU on the VisionAI supports TensorFlow Lite and AutoML Vision Edge frameworks. Imago supplies an SDK and sample programs for image acquisition, I/O handling, and other basic functions. This appears to be Imago’s GUI-based ViewIT framework, which is separately listed on the product page.

ViewIT supports rapid camera and image configuration and application development. The software avoids the need to write code “outside of image processing itself,” says Imago. Sample applications include counting, vibration control, and kinematic monitoring. VisionAI developers can also write their own image processing applications based on Halcon, C++, or Python, “incorporating any libraries or their own source code,” says Imago.

The VisionAI, which is not to be confused with Microsoft’s Vision AI Developer Kit, is equipped with an unnamed 1/1.8”, global shutter CMOS sensor with 5-megapixel (2560 × 1936) resolution and support for frame rates of 65fps. The camera offers a C-Mount or optional Lens protection tube mounting system. An LED strobe capability includes pulse duration sync’d with the shutter, as well as a white LED or optional LED ring illumination function.

The camera integrates 2GB DDR4, a microSD slot for up to 32GB, and a GbE port with cable. The DIO, which is combined in a single cable, includes 2x digital inputs and 4x digital output.

The 50 × 50 × 40.1mm camera weighs 183 grams (including cable) and has a 24VDC (21-28V) input. The VisionAI supports 5 to 40°C temperatures. An optional protective housing can protect connected C-mount lenses.

Imago’s home page also features a white paper (PDF) that explores the event-based vision features of its VisionCam EB camera. The VisionCam EB offers event-based contrast detection CMOS sensors from Prophesee for high-speed tracking. Other features include object counting, vibration measurement, and kinematic monitoring. Imago also manufactures Linux-driven embedded cameras including the VisionCam XM smart area scan camera, which can power an additional GigE camera via a switch, as well as a VisionCam LM line-scan model and a basic VisionSensor PV model.

Further information

No pricing or availability information was provided for the VisionAI. More information may be found on Imago Technologies’ product page.