“Image analysis centered on a centralized control system has its limits. Intellectualizing image analysis and making it in real time at the edge will play an important role from an aspect of improvement in safety and productivity. Intel will provide the best hardware and software platforms for ‘Edge AI’ and we are going to create new opportunities by working with our partners that provide commercialized services.”
Intel Korea held ‘Korea Edge AI Forum’ at KOEX Intercontinental Hotel on the 4th and shared its strategies and portfolios for Edge AI ad its prediction on the industry with its partners.
Intel and its partners all agreed that it is necessary to spread AI workloads to the where data is actually generated. They especially agreed on the need to apply Edge AI to image analysis to extract meaningful data.
“43% of AI-related tasks will be handled by edge computing and there will be 15 times more of devices that can perform AI-related tasks in 2023.” said President Kwon Myeong-sook of Intel Korea. “As things are becoming more dispersed from being centralized, we are currently preparing a portfolio of hardware and software that provide flexibility and expandability as it is difficult to deal with AI workloads with single device.”
Intel provides hardware platforms such as CPU, GPU, FPGA (Field-Programmable Gate Array), and VPU (Vision Processing Unit) for Edge AI. Gen 3 Movidius VPU that is expected to be released during the first half of 2020 has 10 times better inference ability than current products without any increased power consumption. Because Gen 3 Movidius VPU can be used independently unlike Gen 2 Movidius VPU, it can be used for AI accelerator, independent smart camera, drone, and robot. “Because Gen 3 Movidius VPU provides more effective architecture from a standpoint of power consumption and can be used for various form factors, it can provide better performance than high-end GPUs.” said Director Choi Byung-won of Intel Korea.
Intel also provides a software toolkit called ‘OpenVINO’ that is based on convolutional neural networks. OpenVINO helps AI algorithms to be easily optimized for various processors from VPU to Xeon. Intel also introduced ‘DevCloud’ that allows various Edge AI hardware to be tested under the cloud last month.
Various partners of Intel such as LG Electronics, Hanwha Techwin, and Innodep attended the forum and shared AI-related tasks and their examples of collaboration to solve such tasks. LG Electronics is applying deep learning technology to its 3D camera that is being currently developed so that its 3D camera can perform face recognition, body measurement, and obstacle recognition and it has secured compatibility with Intel’s RealSense technology. Hanwha Techwin is planning to optimize software algorithms of its CCTV camera and NVR (Network Video Recorder), which will be released in 2020, through Intel’s OpenVINO toolkit and install Movidius VPU as an accelerator. Innodep is introducing Intel’s technology based on AI to its smart city data management platform.
“While CCTVs in the past recorded videos and saved videos into a storage device for a future use, current CCTVs analyze videos in real time and detect any unusual pattern. CCTVs in the future will be able to predict any unusual pattern beforehand and prevent such pattern from taking place.” said Director Jung Won-seok of Hanwha Techwin. “AI is used for improved quality, smart search, and behavior analysis within the video security industry and it can be used to predict any dangerous situation even without intervention from a person and implement even safer smart city environment.”
Staff Reporter Jung, Hyeonjung | iam@etnews.com