March 4, 2021, China – STMicroelectronics has launched a new AI firmware feature pack and camera module hardware kit, allowing embedded developers to develop edge devices that can run on STM32* microcontroller (MCU)-based edge devices Affordable and powerful computer vision applications.
The STM32Cube feature package FP-AI-VISION1 contains several complete computer vision application code examples that run Convolutional Neural Networks (CNN) on the STM32H747 and can be easily ported across the STM32 full range of products. The firmware presents several application examples where developers can retrain neural networks with selected datasets, providing greater freedom and flexibility for solving various use-case problems.
New features include support for USB VC cameras (webcam mode), simplifying image acquisition tasks, as well as food sorting and user presence detection code samples that create a convenient visual “wake-up” that takes the system from power saving mode wake. In the STM32 Wiki there is an article on how to use the Teachable Machine online tool with the STM32Cube.AI and FP-AI-VISION1 function packages to create an image classification application.
The B-CAMS-OMV camera kit works best with the FP-AI-VISION1 firmware and provides the hardware needed to train and deploy neural network models. The camera kit includes a riser card that houses the STMicroelectronics MB1379 5-megapixel OV5640 color camera module. The riser card is compatible with all STM32 Discovery boards and evaluation boards with ZIF interface, and can also be used with ST’s VG5661 automotive grayscale global shutter camera. In addition, the Waveshare interface and OpenMV interface allow users to connect a variety of third-party infrared and visible light cameras to address a wider range of computer vision applications. There is an article on the STM32 Wiki on how to integrate the code generated by STM32Cube.AI into the OpenMV ecosystem.
FP-AI-VISION1 includes various framebuffer processing functions, camera drivers, as well as image capture software, preprocessing software, and neural network inference software, and several neural network models are available, including floating-point based models and X – Quantized models generated by CUBE-AI, X-CUBE-AI is an artificial neural network C code generator optimized by STMicroelectronics, which allows developers to fine-tune neural models for intended applications because of the support for flexible memory configurations.