Summary of ST’S FP-AI-VISION1 FOR EDGE COMPUTER VISION APPLICATIONS
STMicroelectronics released FP-AI-VISION1, an STM32Cube function pack enabling on-device Edge AI computer-vision on STM32 MCUs (demo on STM32H747). It provides ROM software, image pre-processing (STM32_Image), an optimized neural network runtime (STM32_AI_Runtime via X-CUBE-AI), sample CNN apps (food recognition, person detection), USB webcam support, dataset frame capture, and libraries for testing, debugging, and validation. Compatible across STM32 devices and supports various neural networks with dataset adaptation; sample builds use the STM32H747I-DISCO board with the B-CAMS-OMV camera module.
Parts used in the FP-AI-VISION1:
- STM32H747 microcontroller (STM32H747I-DISCO Discovery board)
- B-CAMS-OMV camera module bundle
- STM32Cube function pack FP-AI-VISION1 (software)
- STM32_Image image pre-processing library
- STM32_AI_Runtime neural network library (generated by X-CUBE-AI)
- X-CUBE-AI Expansion Package for STM32CubeMX
- USB host/host connection for USB webcam application
- Development and debugging libraries for testing, debugging, and validation
Computer vision applications have tremendously increased since the progress in the field of machine learning and artificial intelligence. These applications come at the cost of high-end hardware or cloud support for advanced computations and processing. Hence, STMicroelectronics has announced a new AI software support along with the camera-module hardware, for building affordable applications. The system now supports Edge AI for operating computer-vision applications on the device itself, thus replacing the expensive cloud support for similar high-end applications.
ST’s FP-AI-VISION1 is a STM32Cube function pack that comes with various code instances. This can run computer-vision applications in the environment supporting convolutional neural networks for deep learning on STM32H747. The function pack is flexibly compatible with all STM32 MCUs. Hence, the developers need not start from scratch if they want to switch their STM-based hardware. The firmware also gives an option for using various neural networks with almost every dataset according to the choice of users. Although, the dataset needs to be manipulated so that it is compatible with the neural network.
Features Of ST’s FP-AI-VISION1
- Complete software support on ROM to develop computer vision applications on STM32 microcontroller
- Image pre-processing library, STM32_Image for initial stages of the applications
- Neural Network library optimized for STM32 (STM32_AI_Runtime) generated by means of the X-CUBE-AI Expansion Package for STM32CubeMX
- Food recognition and Person presence detection application based on Convolutional Neural Network
- USB webcam application for computer vision enabling the STM32H747I-DISCO board to capture video when connected to a host’
- It comes with libraries which allow the function such as testing, debugging and validation of the embedded application
- It supports features like camera frame capture to allow image dataset collection, which is a precise way of extracting frames from a video.
- Sample implementations are available for the STM32H747I-DISCO Discovery board connected to the B-CAMS-OMV camera module bundle
Read more: ST’S FP-AI-VISION1 FOR EDGE COMPUTER VISION APPLICATIONS
- What is FP-AI-VISION1?
FP-AI-VISION1 is an STM32Cube function pack by STMicroelectronics that provides software and sample code for running Edge AI computer-vision on STM32 microcontrollers. - Can FP-AI-VISION1 run on different STM32 MCUs?
Yes, the function pack is flexibly compatible with all STM32 MCUs. - Does FP-AI-VISION1 include neural network support?
Yes, it includes the STM32_AI_Runtime neural network library generated via the X-CUBE-AI Expansion Package. - What image processing support is included?
It includes the STM32_Image image pre-processing library for initial stages of applications. - Are there sample applications available?
Yes, sample implementations include food recognition, person presence detection, and a USB webcam application for the STM32H747I-DISCO board with B-CAMS-OMV. - Does FP-AI-VISION1 support dataset collection?
Yes, it supports camera frame capture to allow precise image dataset collection from video frames. - Is cloud computing required for these vision applications?
No, FP-AI-VISION1 enables Edge AI so applications can run on-device without expensive cloud support. - Does the package offer testing and debugging tools?
Yes, it comes with libraries that allow testing, debugging, and validation of embedded applications.
