A few days ago, DJI released the "Xiao" Spark drone, which introduced the face detection and gesture recognition functions for the first time, further consolidating its position in the global drone industry - Apple in the aerial photography industry.
However, speaking of this model, Spark is not a news, because as early as March and April this year, foreign websites have a dismantling diagram of the Spark engineering prototype.
"Xiao" Spark vision chip supplier reveals
In 143 & TImes; 143 & TImes; 55 mm, weight 300G, only the palm-sized drone body contains a core component: Myriad 2 computer vision processing chip, designed by Movidius, now affiliated with Intel Corporation.
Remi El Ouazzane, former CEO of Movidius and current vice president of Intel's new technology, said that DJI has increased its bets in the consumer drone market, and its engineering team has optimized the parts and materials list to try to put many advanced features into it. In a small machine, the Myriad 2 VPU (Visual Processing Unit) chip plays a vital role. It is driven by Movidius 2 to complete functions such as computer vision, deep learning, and image signal processing.
DJI developed the imaging system on Myriad 2 and deployed it to other UAVs. The first cooperation was integrated in Phantom 4 (the LC1860 solution for the chip and the flight control is used), as a tool for sensing and real-time obstacle avoidance. The Movidius chip also allows the drone to hover in a fixed position when it does not receive GPS signals.
Based on the Phantom 4, the DJI team spent about a year developing R&D Spark, using Myriad 2 to make as many advanced visual algorithms as possible and reducing the size and weight. Low power operation.
Movidius is involved in more than twice the workload, mainly computer vision including face recognition, gesture tracking, sensing and real-time obstacle avoidance. There are a large number of internal algorithms based on neural network learning. The goal of DJI is to turn the drone into a "flying AI assistant."
Another feature that Movidius 2 undertakes is Image Signal Processing (ISP). The Spark drone features a professional sports camera-class 1/2.3-inch CMOS sensor and professional aerial lens with an effective pixel count of 12 million. The image processing quality provided by Movidius 2 is equivalent to a stand-alone ISP. At present, there is no commercial chip in the market that can simultaneously load image signal processing and computer vision.
It is reported that Movidius manufactures a system-on-chip (SoC) platform designed to aid computer vision applications, as well as algorithms such as deep learning, navigation, and deep information processing. Other customers include Lenovo, Google and thermal imaging technology company FLIR.
Previously, Dajiang questioned Intel's demonstration of the automatic obstacle avoidance function of the drone, which was suspected of cheating. Intel acquired AscTec, the world's fourth largest drone manufacturer, and did not expect that the two sides would cooperate in such a way.
After being acquired by Intel last September, Movidus officially moved into the Intel Group Building in November. Intel has been promoting its RealSense 3D camera to obtain the three-dimensional geometric information of objects by calculating the positional deviation between the corresponding points of the image. This technology has become a common practice for visual processing of drones.
In recent years, Intel, the CPU giant, has been making acquisitions to accelerate the company's entry into machine learning, deep learning, cognitive computing, autonomous driving and artificial intelligence. Related companies include Nervana Systems, Saffron, Mobileye, etc., to create a leading computer vision processing matrix through a series of initiatives in IP core and innovation.
At present, Movidius's focus is not on the autopilot market, but on surveillance cameras and drone aerial photography. According to EL ouzzane, Movidius's Myriad 2 is currently only a single VPU chip, manufactured by Taiwan's semiconductor foundry giant TSMC.
Their idea is to allow heavy image recognition to be performed on mobile devices, and these features can only rely on the computing power provided by the remote cloud. If you can imagine that it can be achieved, it will bring faster response to mobile devices, and ultimately bring more automated machines. Use Movdidius' low-energy, high-performance SoC chip to capture, navigate, map, and identify scenes and objects, creating intelligent machines with three-dimensional views that understand the state of the environment and navigate accordingly.
With the expansion of functions, Movidius' technology will be integrated into Intel's RealSense depth-sensing camera technology. Currently, this technology has been applied to a webcam on the market, as well as augmented reality head-mounted devices and This technology is also built into the man-machine.
Nanning Ousibang Information Technology Co., Ltd. , https://www.ousibangvape.com