At the Global Consumer Electronics Show (CES 2017), which ended just last weekend, ZEROTECH, the top drone manufacturer from China, understated and mysteriously demonstrated a new prototype of the drone: DOBBY-AI - DOBBY Pocket Drone - AI version. DOBBY-AI is the latest prototype of the smart pocket drone. Like the previously released DOBBY pocket drone, it is only the size of a palm and the weight of the whole machine is equal to one iPhone6 ​​Plus. DOBBY-AI seems to be just a child's toy, and when the presenter let DOBBY-AI really spread its wings and let it fly in the air, it understands our intentions like our partners, not only real-time multi-person detection and tracking, but also Can recognize our various postures.
Since 2012, deep learning has revolutionized artificial intelligence (AI), and unprecedented precision has dominated artificial intelligence applications in areas such as image recognition, speech recognition, and natural language processing. However, have you ever thought about:
1. How to implement complex deep learning algorithms when the device does not have the computing support of the cloud server?
2. How to implement complex deep learning algorithms at very low power consumption and very small size?
The debut of the DOBBY-AI prototype proves that the deep learning solution using FPGA can solve these two problems.
Figure 1 - Gesture recognition effect diagram
Look closely at the DOBBY-AI motherboard: there is a Xilinx fully programmable Zynq? SoC chip. The Zynq All-Programmable SoC family combines the software programmability of the ARM? processor with the hardware programmability of the FPGA to enable critical analysis and hardware acceleration, while also integrating CPU, DSP, ASSP, and hybrid on a single device. Signal function. Undoubtedly, it gives the drone the ability to perceive the surrounding environment, understand the owner's intentions, and help the drone to guarantee the endurance and stability. Burned on top of the Zynq SoC is the Aristotle deep learning processor from Shenjian Technology – which requires only about 3W of power consumption, which enables 230 billion deep learning operations per second.
Figure 2 - Appearance of the DOBBY AI Prototype
In 2017, we will release a new artificial intelligence solution that greatly enriches the way users interact with drones and enhances the drone's user experience. We have worked closely with SZI Technology to develop an intelligent solution based on gesture recognition, which will be one of the core functions of the DOBBY-AI Pocket UAV. —— Zhou Wei, zero-degree intelligent control CTO
Using only a low-end Zynq FPGA platform, Deep Insight's deep learning processor solution supports real-time multi-person detection, gesture recognition, tracking and many other applications. This is the world's first FPGA to be practical on drones. Deep learning program. Because for drone products, deep learning solutions are severely limited by size, power consumption, and price, it is difficult to achieve high processing power while meeting all of these limitations. We are pleased to learn more about the combination of deep compression and Aristotle deep learning processors, and realize the practical use of FPGA deep learning solutions on intelligent drones on Xilinx FPGAs. —— Yao Wei, CEO of Shenjian Technology
Shen Jian Technology is a startup founded by Stanford University and Tsinghua University's world-class deep learning hardware acceleration researchers. Previously, Shen Jian introduced his core technology flow in Hot Chips 2016, and this time, their products were first publicized.
According to Yao Jian, CEO of Shenjian Technology, Shenjian is building a terminal deep learning solution for applications such as drones and security monitoring, as well as cloud deep learning solutions for applications such as speech recognition. The complete set of development tools provided by Shenjian Technology can help users get rid of the trouble of FPGA development. Even if there is no FPGA background algorithm engineer, you can use Shenjian Technology's tool chain to deploy the trained model quickly and efficiently to the hardware.
We believe that Deep Learning's deep learning program has a major impact on the landing of artificial intelligence applications. We can expect FPGAs to give more and more wisdom to every node around us, let us enjoy more colorful life.
Pet Bottle Flake Pelletizing Line
High Capacity Pet Bottle Flake Pelletizing Line,Precision Pet Bottle Flake Pelletizing Line,Affordable Pet Bottle Flake Pelletizing Line,Customizable Pet Bottle Flake Pelletizing Line
Zhejiang IET Intelligent Equipment Manufacturing Co.,Ltd , https://www.ietmachinery.com