OpenMV has launched two new edge AI camera boards programmable with MicroPython: the OpenMV AE3 powered by an Alif Ensemble E3 dual Cortex-M55, dual Ethos-U55 micro NPU SoC, and the larger OpenMV N6 board based on an STMicro STM32N6 Cortex-M33 microcontroller with a 1 GHz Neural-ART AI/ML accelerator. Both can run machine vision workloads for several years on a single battery charge.
The OpenMV team has made several MCU-based camera boards and corresponding OpenMV firmware for computer vision, and we first noticed the company when they launched the STM32F427-based OpenMV Cam back in 2015. A lot of progress has been made over the years in terms of hardware, firmware, and software, but the inclusion of AI accelerators inside microcontrollers provides a leap in performance, and the new OpenMV N6 and AE3 are more than 100x faster than previous OpenMV Cams for AI workloads. For example, users can now run object detection models, track human body pose, see facial landmarks, and more at 30 FPS with microcontroller-class hardware.
OpenMV AE3
Specifications:
- MCU – Alif Ensemble E3
- Dual-core CPU
- Arm Cortex-M55 core @ 400 MHz
- Arm Cortex-M55 core @ 160 MHz
- Dual AI accelerator
- Arm Ethos-U55 microNPU @ 400 MHz with 256 MAC/c (204 GOPs)
- Arm Ethos-U55 microNPU @ 160 MHz with 128MAC/c (46 GOPs)
- Both NPUs can be used simultaneously
- 2D GPU for image scaling
- Memory – 13.5MB SRAM
- Dual-core CPU
- Storage – 32MB flash (200MB/s)
- Camera – 1MP color global shutter camera; 120 FPS at VGA resolution
- Audio – Built-in microphone
- Wireless – 2.4 GHz WiFi 4 and Bluetooth LE 5.1
- USB – 1x USB-C port (480 Mbps)
- Sensors
- 8×8 ToF sensor with 4-meter range
- 6-axis IMU sensor (accelerator and gyroscope)
- Expansion
- Qwiic connector
- 7 through holes with GPIO, 3.3V, GND
- 10-pin GPIO B2B header on the back of the board
- Debugging – SWD via Edge Connector
- Misc
- User Button
- User RGB LED
- Power Supply – 5V via USB-C
- Power Consumption (5V)
- 60mA (0.25W) when running YOLO at 30 FPS
- About 30mA (0.12W) while idling
- Under 500uA (2.5mW) in deep-sleep mode for years of battery life while waking up on sound, motion, and a date/time.
- Dimensions – 2.54 x 2.54 cm (board only)
The OpenMV AE3 can run object detection models like YOLO at ~30 FPS while drawing only 0.25W. The company also says the board idles at 0.12W and consumes only 2.5mW in deep sleep, which in practical terms means that three AA batteries can power the board for more than a day at full power, nearly three days idling, and more than four months in deep sleep. They expect to further reduce power consumption with optimization with the goal of reaching <0.25mW draw in deep sleep.
OpenMV N6
Specifications:
- MCU – STMicro STM32N6
- MCU Core – Arm 32-bit Cortex-M55 CPU @ up to 800MHz with Arm Helium and Arm MVE
- GPU – Neo-Chrom 2.5D GPU, Chrom-ART Accelerator (DMA2D)
- NPU – ST Neural-ART Accelerator @ 1 GHz, 600 GOPS
- BPU – Hardware accelerated H.264 and JPEG encoders
- Memory – 4.2MB SRAM
- Memory – 64MB PSRAM (800 MB/s)
- Storage
- 32MB flash (200MB/s)
- MicroSD card slot on the bottom side of the board
- Camera
- Replaceable 1MP color global shutter camera; 120 FPS at VGA resolution
- Support for up to 5MP camera sensors,
- Audio – Built-in microphone
- Networking
- Gigabit Ethernet PHY (but no RJ45 jack on board, expansion board needed)
- 2.4 GHz WiFi 4 and Bluetooth LE 5.1
- USB – 1x USB-C port (480 Mbps)
- Sensor – 6-axis IMU sensor (accelerator and gyroscope)
- Expansion – 2x 16-pin headers with 18x GPIOs
- Debugging – 10-pin JTAG & SWD connector
- Misc
- User Button
- User RGB LED
- Power Supply
- 5V via USB-C
- 2-pin connector for 3.7V LiPo battery; charging circuitry
- Power Consumption – Under 0.75 Watts
- Dimensions – Still small, but quite bigger than the OpenMV AE3…
The OpenMV N6 is more powerful than the smaller OpenMV AE3, but consumes more power under load. The STM32N6 board supports up to 5 MP cameras, and the on-chip hardware H.264 encoder allows for the recording of MP4 videos on the microSD card slot, or streaming via WiFi/BLE or Gigabit Ethernet.
Software and applications
The OpenMV N6 and AE3 AI camera boards come both preloaded with MicroPython firmware to handle MQTT or API messages and are controlled through Python 3 scripts from a host. The easiest way to program and control the cameras is through the OpenMV IDE available for Windows, Mac, Linux, and even Arm-based platforms like the Raspberry Pi.
Users can easily load AI models, run scripts, check the output from the OpenMV N6/AE3 board, etc… The OpenMV IDE has a full Python Language Server and GitHub CoPilot Integration to help users write code. The firmware and debug protocol are open-source and you’ll find the source code on GitHub.
The OpenMV cameras can be integrated into robots, for example, to track a ball in a competition, a battery-powered wildlife camera that only captures an image when animals (and only animals) are detected, read gauges and meters remotely in a factory, automate drone landing on a target, track baby chicks hatching using thermal vision (the OpenMV N6 can take thermal cameras too), etc…
You can check the video embedded below to have a better idea of what’s possible with the OpenMV AI camera boards.
The OpenMV N6 and AE3 low-power edge AI cameras have just been launched on Kickstarter with a $50,000 funding target. Rewards start at $50 for the AE3 with an enclosure and a 3-month Roboflow Basic plan, while the N6 board goes for $100 with the same plan. Those are early bird pledges for the first 50 units, and “normal” prices are respectively $80 and $120. Shipping adds $20 for worldwide shipping. Deliveries are scheduled to start in December 2025, so you’d have to be patient.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
