| 韦世玮

Edit | Shi Yaqiong

**

36 I learned that recently, “I know the future” announced that tens of millions of yuan A round financing, by the morning leader, Amber, Guo Hongjiaxin, and Miba capital served as long-term financing financial advisers. It is understood that this round of funds will be mainly used for market expansion and development of new products.

I know that in 2017, I have been based primarily based on computer vision technology, deep learning convolutional neural network, automatic machine learning, neural network search, etc., for extremely weak light scene, high-speed sports scene, complex ray scene, etc., etc., etc. The module, realizes high-quality real, full color image reconstruction, its products and systems are mainly used in drone, automatic driving, security, smart mines and other fields.

Taking the extreme weak light imaging of AI-ISP as an example, the lowest brightness of the night vision scene is 0.001LUX compared to the day, and the 100000LUX light weight of the day is 8 orders of magnitude, so the visual information that can be present at night is very small. Many equipment’s night imaging is very difficult.

36氪首发|「深知未来」完成数千万A轮融资,全彩夜视产品已落地多个全天候安防场景

For a long time, the mainstream night vision technology is divided into two paths: one is thermal imaging, mainly converting heat into electrical signals by infrared detectors, but the technology cannot present details, high for scene requirements; second, photon detector, utilization The optical principle enlarges weak light and image, increasing brightness and contrast, but the cost is high, which is very dependent on the imported optic device.

36氪首发|「深知未来」完成数千万A轮融资,全彩夜视产品已落地多个全天候安防场景

In addition, traditional night vision technology also has monochrome images, low resolution, fear of strong light, high prices. From a component perspective, there is currently no low-illuminating CMOS to adapt to the very low darkness at night, and the Sensor vendor is only 0.1lux, while traditional algorithms and ISP vendors are difficult to handle polar signal-to-noise ratio.

In the best of the future joint founder & CEO Zhang Qining, AI night vision technology is the only technical path that can realize the full color night vision, with full-color imaging, high resolution, not fear of strong light, low cost.

The reason is that the visible optical band of human naked eye is 380 nm ~ 790 nm, accounting for only one billion spectrum, and traditional night vision technology uses the infrared spectrum of the naked eye to imaging, and Ai technology can directly enter the visible optical band. Restore color.

At present, there are two main types of products in the future, one is to provide the night vision core form, including Xuan Moon D1, Xuan Moon X4; Second, direct delivery of the complete night vision products, such as Xuan Moon p5 portable camera, night vision Cloud tank machine, single-grade / bienger video machine, etc. Among them, the future night vision drone camera movement can be used to depth with PTZ, which is applied to industrial drones.

36氪首发|「深知未来」完成数千万A轮融资,全彩夜视产品已落地多个全天候安防场景

From a technical point of view, the future advantage is the innovation of the ISP architecture. As a collection of image processing algorithms, ISP is responsible for completing the conversion from Sensor RAW to image data, and the number of corrections, denoising, conversion, and enhancement, etc., including 2000 to 3,000 parameters.

This traditional ISP processing path is a serial mechanism, and the error is easily enlarged, which means that if the initial model is not accurate enough, then this error will be enlarged in a step in subsequent link, which ultimately leads to image distortion. At the same time, traditional ISPs often face problems such as light and dark, camera jitter.

36氪首发|「深知未来」完成数千万A轮融资,全彩夜视产品已落地多个全天候安防场景

A well-known future AI ISP refactors the algorithm architecture of ISP by simulating visual imaging. Based on the depth neural network, it is well known that a model containing hundreds of millions of parameters can improve the stability and adaptability of the model, thereby solving complex rays, jitter and frame rates. At present, it is well known that the AI ​​ISP can increase the brightness gain over 7000 times in 1/30 seconds, and the minimum illumination of hardware is raised from 0.1 lib to 0.003LUX.

From the perspective of the industry, there are currently long-term layouts and technological accumulation in the night vision applications. Zhang Qin tells 36, now Haikang, the minimum illuminance of the night vision products of Dahua can achieve 0.1lux, mainly the full moon scene, and Sony has increased the hardware’s minimum illuminance to 0.07LUX However, but the image is still noise, and the AI ​​application of the upper layer is difficult to structure.

Among them, in the first quarter of 2020, Sony’s CMOS sales have exceeded 350 million, and the city share is 24.7%; Haowa’s sales is 1.7 billion, and the market share is 12.3%.

Overall, the application scenarios of night vision technologies are wide, including 14 TO B scenes, oil field monitoring, marine oil well platform monitoring, port monitoring, tunnel monitoring, night delivery, water conservancy monitoring, railway monitoring, but these scenes still It mainly uses thermal imaging techniques, so it is known that the expansion space in the future AI ISP technology is large.

In the past year, the future of the full color night vision products have been landing many all-weather security scenarios. Among them, in China’s oil cooperation, I know that this year will continue to expand 10-20 new projects. It is expected that the project amount has more than 10 million. The company’s goal has completed the entire scene of China-oil customers in the next three years to become the first A large security night vision solution supplier.

Zhang Qining talked about that security is expected to be a large-scale industry in the company next year. At the same time, in addition to the TO B field, it is well known that in the future will further expand in the car, medical and consumer scene. Among them, according to the prospective industrial research institute, it is expected that in 2025, my country’s car camera market is expected to exceed 23 billion RMB. With the high-end automatic driving of domestic and foreign, the car camera will usher in a billion market.

The team level, knowing the future foundation team by Tencent, Baidu, Xiaomi, Huawei, the algorithm expert from the industry and other industries and product experts. Reflections the future joint founder & CEO Zhang Xi Ning graduated from Wuhan University Computer Department, former Huawei, Tencent Technical Experts and product experts, in Tencent Innovation Center, Tencent Research Institute, IEG and other departments have served as R & D, planning and management positions; knowing the future Joint Foundation & CTO Guo Qiqi once served by Tencent SoSO Search System Architect, Baidu Business Search Department Basic Retrieval Group Technical Leader, Xiaomi MiUi9 Multiple Artificial Intelligent Project Technical Legal Intelligent Project Technology.

Investors said:

Yuanwang capital investor point

Dashen investment investor point

Amber capital investor view

Guo Hongjiaxin investor point