Introducing the OpenMV AE3 and N6 Cameras

The OpenMV Camera

For over a decade now, OpenMV has enabled developers to easily add computer vision to their projects, products, and services. There have been several generations of the namesake OpenMV Cam released, with increasing performance and capabilities as new and more powerful microcontrollers get built by chip makers. These devices are a popular choice in the developer community due to the flexibility and ease-of-use of the boards, GPIO expansion, MicroPython programmability, superb IDE, as well as being open source and very thoroughly documented. Developers also have a range of options available when it comes to the type of computer vision they can make use of, thanks to interchangeable sensors, lenses, and the different MCU features across the product lineup.

However, one feature that was not previously available in any of the OpenMV devices was a dedicated AI accelerator, as none of the models were based on MCU's with built-in neural processing units (NPUs). Native AI accleration allows for high-speed machine learning tasks like object detection, image classification, and semantic segmentation for Edge AI applications.

That changes today though, with the launch of not one, but two new AI-accelerated OpenMV Cam devices on Kickstarter!

OpenMV AE3

OpenMV cameras have always been small, low-power devices thanks to being built around microcontrollers, but the new AE3 is an even tinier, cheaper, and more efficient OpenMV device than ever before.

At the heart of the new OpenMV board is an MCU from Alif Semiconductor, called the E3. It is part of the Ensemble family, and it contains two Arm Cortex-M55 MCU cores and two Arm Ethos-U55 microNPU cores. The primary MCU core runs at 400MHz, while the second core runs at 160MHz. The primary NPU core has 256 MACs (multiply-accumulate), while the second Ethos core is a 128 MAC unit. Combined, the processor has a total of 250 GFLOPS of compute power, or roughly double that of a Raspberry Pi 5 - all while running on only 30 milliamps of power at 5V, and in a package that is only 6mm x 7mm!

The E3 also contains 13.5MB of 3GB/s SRAM on-chip for the application code, with a 200 MB/s OctalSPI link to a secondary 32MB of memory where AI models are stored and executed by the microNPU.

More detailed information on the Alif Ensemble can be found here.

A new OpenMV layout

Previous OpenMV Cams have all followed the same form factor, with only the USB port being updated and one versus two rows of GPIO, but the new AE3 introduces a new, smaller size option for OpenMV users. Measuring only one inch by one inch, the AE3 is TINY. Roughly the size of a postage stamp or a coin, the AE3 can fit nearly anywhere. The board includes Wi-Fi and Bluetooth for connectivity (with a tuned antenna for greater range), a USB-C port to connect to a PC, and a board-to-board connector on the back for future expansion capabilities and add-ons. There is also a microphone, an accelerometer, a Time of Flight (ToF) sensor, and an RGB LED. For additional sensor expansion, there is a Qwiic connector that allows an entire ecosystem of SparkFun sensors and accessories to be added. The GPIO carry PWM, UART, I2C, I3C, SPI, and CAN for added peripheral connectivity.

The camera sensor included on the board is a 1MP color camera, with a global shutter capable of 30 frames per second at 1280x800, and 120 FPS at 640x480. With the NPU acceleration from the Alif Ensemble E3, YOLO models can be run at 25 FPS, again keeping in mind that this device is running on only 60 milliamps of power at 5V! The board can also go down to about 500 microamps when in sleep mode, preserving battery life.

The OpenMV N6

Also launching today is a new member of the OpenMV family that aligns to the traditional size and shape of the existing OpenMV cameras, yet also provides for a massive uplift in performance again due to the inclusion of a dedicated AI acclerator. The OpenMV N6 is based on the STMicroelectronics STM32N6, which is powered by an Arm Cortex-M55 running at 800MHz, and the ST Neural-ART accelerator™, which is a 600 GFLOPS NPU for computer vision tasks. The board also features a whopping 64MB of RAM, running at 800MB/s, and 32MB of Flash storage with a 400 MB/s interface. With all of this processing power and speed, YOLO computer vision models run at a full 30fps!

As this board is larger than the AE3, it has more room for expansion and customization, with 18 GPIO pins carrying PWM, UART, I2C, I3C, SPI, and CAN, as well as interchangeable camera lenses and sensors, allowing for maximum flexibility. The board also includes Wi-Fi and Bluetooth for connectivity, has a microphone, RGB LED, accelerometer and gyroscope, user buttons, and a UHS-1 speed SD card slot for expanded storage. The STM32N6 is unique in that it includes h.264 video/JPEG encode capability in hardware, and has a 10/100/1000 (gigabit) Ethernet PHY built-in. There is even a LiPo battery charger integrated into the board, supporting 3.7 volt batteries.

The camera sensor options are diverse, with support for a 1MP 120/240/480fps color global shutter camera, a 5MP 30fps HDR camera, a FLIR Boson or Lepton thermal camera (or combo thermal + RGB),

There is also the option to add the newly released Prophesee GENX320 event camera that can be used to visually identify changes in light intensity, which is a novel method for mimicking the human eye and how it interprets light:

These added features do of course require more power than the AE3, but the STM32N6 is still an MCU — not a CPU — so even with all of these features and functionality, the system only consumes about 180mA at 5V at full power.

Like the AE3, there is a deep sleep state, where the device can drop down to only about 1mA or so when not in an active state, which dramatically extends battery life.

Use cases

OpenMV cameras have always been extremely flexible in the types of applications that can be built and run, thanks to their MicroPython programming and large ecosystem of add-on boards, sensors, camera options, and GPIO. The new OpenMV AE3 and N6 are no different, and should be useful in projects ranging from industrial, environmental, retail, and medical, to home automation and more. Human presence detection and tracking, facial landmark identification, pose estimation, gesture recognition, audio classification, and motion anomaly detection are just some of the possible use cases that come to mind.

Of course, you can also try out the many existing OpenMV projects that are already posted on the OpenMV website and blog, or here in the Hackster OpenMV Hub. For example, this predictive maintenance vehicle counter looks awesome!

Where to buy the AE3 and N6

The OpenMV Cam AE3 and N6 have launched on Kickstarter today. This is the third Kickstarter campaign run by the team, and with the wild success of the first two campaigns, this one will likely be a smash hit as well — especially considering the small size, high performance, native AI capability, and low price point.

Finally, don't miss Kwabena Agyeman on Hackster Café with Alex Glow as they dive into the new boards and talk about the exciting features, and discuss awesome projects waiting to be built by the community!