The OpenMV Digicam
For over a decade now, OpenMV has enabled builders to simply add pc imaginative and prescient to their tasks, merchandise, and providers. There have been a number of generations of the namesake OpenMV Cam launched, with growing efficiency and capabilities as new and extra highly effective microcontrollers get constructed by chip makers. These units are a well-liked selection within the developer group because of the flexibility and ease-of-use of the boards, GPIO enlargement, MicroPython programmability, very good IDE, in addition to being open supply and really completely documented. Builders even have a variety of choices obtainable in the case of the sort of pc imaginative and prescient they’ll make use of, because of interchangeable sensors, lenses, and the totally different MCU options throughout the product lineup.
Nevertheless, one function that was not beforehand obtainable in any of the OpenMV units was a devoted AI accelerator, as not one of the fashions had been primarily based on MCU’s with built-in neural processing models (NPUs). Native AI accleration permits for high-speed machine studying duties like object detection, picture classification, and semantic segmentation for Edge AI functions.
That adjustments in the present day although, with the launch of not one, however two new AI-accelerated OpenMV Cam units on Kickstarter!
OpenMV AE3
OpenMV cameras have all the time been small, low-power units because of being constructed round microcontrollers, however the brand new AE3 is an excellent tinier, cheaper, and extra environment friendly OpenMV gadget than ever earlier than.
On the coronary heart of the brand new OpenMV board is an MCU from Alif Semiconductor, referred to as the E3. It’s a part of the Ensemble household, and it incorporates two Arm Cortex-M55 MCU cores and two Arm Ethos-U55 microNPU cores. The first MCU core runs at 400MHz, whereas the second core runs at 160MHz. The first NPU core has 256 MACs (multiply-accumulate), whereas the second Ethos core is a 128 MAC unit. Mixed, the processor has a complete of 250 GFLOPS of compute energy, or roughly double that of a Raspberry Pi 5 – all whereas operating on solely 30 milliamps of energy at 5V, and in a bundle that’s solely 6mm x 7mm!
The E3 additionally incorporates 13.5MB of 3GB/s SRAM on-chip for the appliance code, with a 200 MB/s OctalSPI hyperlink to a secondary 32MB of reminiscence the place AI fashions are saved and executed by the microNPU.
(📷: Alif Semiconductor)
Extra detailed info on the Alif Ensemble may be discovered right here.
A brand new OpenMV format
Earlier OpenMV Cams have all adopted the identical type issue, with solely the USB port being up to date and one versus two rows of GPIO, however the brand new AE3 introduces a brand new, smaller dimension choice for OpenMV customers. Measuring just one inch by one inch, the AE3 is TINY. Roughly the dimensions of a postage stamp or a coin, the AE3 can match almost anyplace. The board contains Wi-Fi and Bluetooth for connectivity (with a tuned antenna for larger vary), a USB-C port to connect with a PC, and a board-to-board connector on the again for future enlargement capabilities and add-ons. There may be additionally a microphone, an accelerometer, a Time of Flight (ToF) sensor, and an RGB LED. For added sensor enlargement, there’s a Qwiic connector that enables a whole ecosystem of SparkFun sensors and equipment to be added. The GPIO carry PWM, UART, I2C, I3C, SPI, and CAN for added peripheral connectivity.
The digicam sensor included on the board is a 1MP colour digicam, with a worldwide shutter able to 30 frames per second at 1280×800, and 120 FPS at 640×480. With the NPU acceleration from the Alif Ensemble E3, YOLO fashions may be run at 25 FPS, once more protecting in thoughts that this gadget is operating on solely 60 milliamps of energy at 5V! The board can even go all the way down to about 500 microamps when in sleep mode, preserving battery life.
The OpenMV N6
Additionally launching in the present day is a brand new member of the OpenMV household that aligns to the normal dimension and form of the prevailing OpenMV cameras, but additionally supplies for a large uplift in efficiency once more because of the inclusion of a devoted AI acclerator. The OpenMV N6 is primarily based on the STMicroelectronics STM32N6, which is powered by an Arm Cortex-M55 operating at 800MHz, and the ST Neural-ART accelerator™, which is a 600 GFLOPS NPU for pc imaginative and prescient duties. The board additionally includes a whopping 64MB of RAM, operating at 800MB/s, and 32MB of Flash storage with a 400 MB/s interface. With all of this processing energy and velocity, YOLO pc imaginative and prescient fashions run at a full 30fps!
As this board is bigger than the AE3, it has extra room for enlargement and customization, with 18 GPIO pins carrying PWM, UART, I2C, I3C, SPI, and CAN, in addition to interchangeable digicam lenses and sensors, permitting for max flexibility. The board additionally contains Wi–Fi and Bluetooth for connectivity, has a microphone, RGB LED, accelerometer and gyroscope, person buttons, and a UHS-1 velocity SD card slot for expanded storage. The STM32N6 is exclusive in that it contains h.264 video/JPEG encode functionality in {hardware}, and has a ten/100/1000 (gigabit) Ethernet PHY built-in. There may be even a LiPo battery charger built-in into the board, supporting 3.7 volt batteries.
The digicam sensor choices are various, with help for a 1MP 120/240/480fps colour international shutter digicam, a 5MP 30fps HDR digicam, a FLIR Boson or Lepton thermal digicam (or combo thermal + RGB),
There may be additionally the choice so as to add the newly launched Prophesee GENX320 occasion digicam that can be utilized to visually determine adjustments in mild depth, which is a novel technique for mimicking the human eye and the way it interprets mild:
These added options do after all require extra energy than the AE3, however the STM32N6 continues to be an MCU — not a CPU — so even with all of those options and performance, the system solely consumes about 180mA at 5V at full energy.
Just like the AE3, there’s a deep sleep state, the place the gadget can drop all the way down to solely about 1mA or so when not in an energetic state, which dramatically extends battery life.
Use circumstances
OpenMV cameras have all the time been extraordinarily versatile within the forms of functions that may be constructed and run, because of their MicroPython programming and enormous ecosystem of add-on boards, sensors, digicam choices, and GPIO. The brand new OpenMV AE3 and N6 aren’t any totally different, and must be helpful in tasks starting from industrial, environmental, retail, and medical, to dwelling automation and extra. Human presence detection and monitoring, facial landmark identification, pose estimation, gesture recognition, audio classification, and movement anomaly detection are simply a number of the attainable use circumstances that come to thoughts.
After all, you may as well check out the numerous present OpenMV tasks which might be already posted on the OpenMV web site and weblog, or right here within the Hackster OpenMV Hub. For instance, this predictive upkeep car counter appears superior!
The place to purchase the AE3 and N6
The OpenMV Cam AE3 and N6 have launched on Kickstarter in the present day. That is the third Kickstarter marketing campaign run by the crew, and with the wild success of the primary two campaigns, this one will seemingly be a smash hit as nicely — particularly contemplating the small dimension, excessive efficiency, native AI functionality, and low value level.
Lastly, do not miss Hackster Café on March 18 at 10:30am PST, with Alex Glow and Kwabena Agyeman as they dive into the brand new boards and speak in regards to the thrilling options, and focus on superior tasks ready to be constructed by the group!
