The OpenMV N6 is the first python-powered machine vision module capable of running a Yolov8n image detection model at more than 30 frames per second at 256×256 while needing less than 0.75W, thanks to its STM32N6. And to bring this to more engineers, OpenMV, an Authorized ST Partner, is launching a Kickstarter campaign, thus enabling early adopters to get the first units at a discount while also participating in its launch. Developers also get to be the first to work on new application examples and implementations from OpenMV, which showcase on-device processing and the ability to rapidly tweak applications running on the STM32N6 thanks to its IDE and MicroPython framework. Backers can pre-order the OpenMV N6 for USD120.
The OpenMV N6 features a 1-megapixel global shutter camera capable of recording at up to 120 frames per second. It also comes with a 2.4 GHz Wi-Fi b/g/n controller, Bluetooth LE 5.1 module, and a Gigabit Ethernet device that requires external shield. In fact, OpenMV even offers a Power over Ethernet daughterboard and a 3.7V LiPO charger port, so the OpenMV N6 can be more mobile and practical. The new system also features a USC-C High-Speed port capable of 480 Mb/s, 32 MB of Flash at 400 MB/s, and 64 MB of PSRAM at 800 MB/s. OpenMV N6 even supports JTAG and devices like STLINK to make programming it easier.
Where is computer vision today?
The existing OpenMV landscape
The OpenMV cameras have been quite popular, with more than 100,000 units sold, many of them featuring STM32 devices like the STM32F4 or the STM32F7. As of today, OpenMV sells two other camera modules featuring an STM32, the OpenMV Cam R2 and the OpenMV Cam H7 Plus. Both are descendants of the OpenMV Cam H7, which also launched with a Kickstarter campaign in 2019. Backers will thus find comfort in the fact that OpenMV has significant crowdfunding experience, which goes a long way in mitigating the inherent risks that such an operation entails.
On our end, the ST Blog sat down with Kwabena Agyeman, President and Co-Founder at OpenMV, to understand why the company is announcing the OpenMV N6 only a few months after the mass-market launch of the STM32N6, our most powerful STM32 MCU and the first with an NPU. The company does have extensive experience with STM32 microcontrollers, which partly explains why it can come up with a new design on a recently released MCU so quickly. In many instances, the work that OpenMV did on other STM32 devices carried over to this new system. Additionally, it is interesting to see OpenMV jump so quickly on the new device to enable new applications.
The new era of the STM32N6
As Agyeman explained, OpenMV is increasingly attracting engineers working on industrial applications, thus demonstrating a new blurring of the lines between enthusiasts, prosumers, and niche applications. Computer vision is here to stay. Myriad of smart factories, businesses, and professionals with arcane needs are clamoring for it. However, massive installations with expensive graphics cards, server rooms, and dedicated electrical wiring are often not feasible. OpenMV N6 enables applications previously impossible: mainly industrial-grade computer vision running locally with something as unobtrusive as Power Over Ethernet or a battery. Additionally, thanks to its use of MicroPython, it can bring industrial applications to all engineers, regardless of their experience level.
What made the OpenMV N6 possible?
Performance
The OpenMV co-founder explained that the STM32N6 opens the way for entirely new computer vision applications on a microcontroller, but it’s not always for the reason most people think of. Indeed, many will assume that the neural processing engine, called Neural-ART Accelerator, is solely responsible for this feat. And it does indeed play a significant role thanks to its 600 GOPS, which makes it 600 times more powerful in AI tasks than our fastest microcontroller without an NPU. Thanks to its two 64-bit AXI memory buses, the STM32N6 can process the vast amount of data that passes through the neural network when running machine learning algorithms.
However, Agyeman also explained that the large 4.2 MB of RAM, the camera pipeline, and the JPEG encoder and decoder greatly impact performance to enable applications that could not run on an OpenMV module before because it would have required too much processing and electrical power. Indeed, this exact specification makes the STM32N6 pretty unique in the industry, which is why we are also offering a version of the device without the Neural-ART Accelerator for engineers only looking for large memory and multimedia capabilities. Thanks to its Cortex-M55 at 800 MHz, two I3C interfaces, two ADCs, and many 32-bit advanced timers, the STM32N6 is our most powerful STM32 yet and is uniquely positioned for many applications.
Efficiency
Besides performance, the STM32N6 also brings a level of power efficiency that is critical for the success of the OpenMV N6. Indeed, thanks to its tight consumption and low power modes, it’s possible to run an application on the STM32N6 by using a battery or Open MV’s Power over Ethernet shield. For instance, thanks to the STM32N6 deep sleep mode, it’s possible to use the OpenMV N6 on a LiPo battery for over a year before replacing it. Moreover, it doesn’t require a heatsink or fan, making it even more practical. For instance, the lack of active cooling means that designing a case for the OpenMV N6 and using a 3D printer to make it is vastly simpler.
What’s next?
The ability to rapidly prototype and run proofs-of-concept on the OpenMV N6 is another reason many find the module attractive. The fact that OpenMV heavily relies on MicroPython vastly simplifies developments. Teams don’t have to compile their firmware after every little change, and it hides a lot of the underlying complexity, thus lowering the barrier to entry. Moreover, onboard computer vision has significant privacy benefits since images stay on the board, and developers don’t need to rely on clouds and data transfer over the Internet. Hence, we expect systems like OpenMV N6 to help democratize computer vision as more industries depend on it.
