비전 AI 디바이스가 △자동차 △보안 △제조 등 다양한 산업군에서 채택되면서 개발단계서부터 빠른 시장 대응이 필요해지고 있다. 이에 개발 프로세스 단축 효과를 주면서 동시에 높은 성능과 신뢰성을 가진 비전 AI 스타터 기트가 하나의 선택지로 부상하고 있다.
ATUS Collaboration Develops Event-Based Vision Sensor Platform
Processing pixels with different brightness levels and processing data compared to the original ↓
▲ KV260 Vision AI Starter Kit (Photo: AMD)
As vision AI devices are adopted in various industries such as automobiles, security, and manufacturing, rapid market response from the development stage is becoming necessary. Accordingly, a vision AI starter kit that shortens the development process while also providing high performance and reliability is emerging as an option.
AMD announced on the 24th that it has developed an 'Event-based Vision Sensor (EVS) development platform' based on its Kria KV260 Vision AI Starter Kit in collaboration with ATUS, a domestic digital media and embedded FPGA development company.
Event-based vision sensors are devices that detect changes in brightness in an image, processing only those pixels in the image acquired through a camera where changes in brightness occur. Therefore, it can reduce the amount of data processing compared to existing image sensors that perform calculations based on frames, and has advantages such as △fast process and response speed △low power consumption, etc.
AMD and ATUS announced that they plan to continue their collaboration in the future to supply products to domestic and international markets in various fields such as automotive, security, manufacturing, and medical.
This development platform designed by ATUS features the IMX637 event-based vision sensor from Sony Semiconductor Solutions on AMD’s Cria KV260 Vision AI Starter Kit. It utilizes FireNet as a neural network for reconstructing acquired image data, and provides fast computation and low-latency performance capable of processing images of 640x480 pixels (VGA).
Based on these strengths, event-based vision sensors can be used in a variety of areas, including driver monitoring in automobiles, inspection of production lines in manufacturing facilities, environmental recognition through unmanned systems, and counting the number of visitors to retail stores.
In addition, it was emphasized that it is suitable for medical environments such as hospitals or home care because it does not output sensitive personal information-related data such as detailed features or shapes of people or objects or color information.
Meanwhile, AMD’s Kria KV260 Vision AI Starter Kit is a development platform for the Kria K26 System-on-Module (SOM), part of the Kria SOM portfolio of small form factor, production-ready embedded boards that enable rapid deployment of edge-based applications.
It is a development platform that supports vision-accelerated applications provided by the AMD Xilinx App Store, and is known to support economical accelerated application design as developers can get it up and running in less than an hour without hardware knowledge.