Image source @ Visual China
At a time when the AI track is being questioned and many companies continue to lose money, since late March, a number of 3D vision companies have announced financing news:
On March 23, Transdimensional Intelligence, founded in June 2021, announced the completion of nearly 10 million US dollars of angel round financing; on March 24, Landmark Technology, which was established in July 2021, announced the completion of tens of millions of yuan of angel round financing; on March 29, Yishi Technology, which has been focusing on the field of industrial 3D vision for 5 years, announced the completion of more than 100 million yuan A round and A+ round of financing.
According to reports, cross-dimensional intelligence provides 3D vision cameras, algorithms and software and hardware integration and other three-dimensional vision field solutions, has been applied to semiconductors, 3C electronics, automotive, metal processing and other industrial manufacturing scenarios; Landmark Technology provides 3D disorderly grasping system, 3D point cloud trajectory guidance positioning system and 3D point cloud high-precision detection system; Yishi Technology's products also cover the consumer electronics, automotive and new energy fields of 3D detection and measurement market.
It can be seen that the automotive, industrial and other fields are still the main application scenarios of 3D vision new financing enterprises. Of course, there are also some mobile phone brands that use 3D deep-sense cameras, such as the latest Honor Magic4 Pro, iPhone 13 series, etc.
In the AI track, computer vision is the largest branch, accounting for more than half of the market share. According to statistics, in 2019, the scale of the core industries of computer vision and the scale of related industries in the mainland were 63.33 billion yuan and 143.86 billion yuan respectively, accounting for 58.2% and 37.6% of the AI core and related industries, respectively.
The biggest application scenario of computer vision is security imaging. According to statistics, in 2020, security still occupies 67.9% of the computer vision application layer, and the application technology is still dominated by 2D vision. The application of 3D vision in the field of industry and other fields, placed in the sea of computer vision, can only be regarded as a small pond.
But in 2022, that could be about to change.
iPhone X and the first wave of 3D machine vision entrepreneurship in China
From black and white to color, from low resolution to high resolution, from still images to moving images, from 2D planes to 3D stereoscopic, human visual sensing technology has undergone four revolutions. Strictly speaking, the fourth visual revolution represented by 3D vision technology is fundamentally different from the first three: the first three are based on 2D planes; in the field of 3D vision, there is no difference between black and white and color, and even whether there is ambient light does not affect the construction of three-dimensional information.
The four visual revolutions are distinguished by technical type, and the relationship is not progressive in time. In fact, as early as 1922, the first 3D film "The Power of Love" was shot with two cameras shooting red and green pictures. But it wasn't until September 2017, when Apple released the iPhone X, which was equipped with a 3D structured light depth camera for the first time, that 3D vision technology detonated public awareness for the first time, and even triggered a big discussion about whether mobile phones should be based on fingerprint recognition or 3D face recognition.
But this big discussion in the technology media only involves the significance of the first level of Apple's introduction of 3D structured light technology - it can be quickly unlocked without human contact. In fact, Apple, based on 3D sensors, also wants to further develop other applications and services such as augmented reality (AR). Unlocking is only the first step.
The 3D laser scanning of the iPhone X uses VCSEL (Vertical Cavity Surface Emitting Laser) laser, which has strong advantages in terms of accuracy, miniaturization, low power consumption and reliability compared to LED infrared light sources. Previously, 3D vision was mainly applied in a small number of applications in the industrial field, and the entire market size was not large, including upstream suppliers of various types of 3D vision sensors, midstream 3D vision perception solution providers, and downstream application algorithm solution providers.
In the field of 3D vision, laboratory products have appeared as early as the 1980s, and industrial and consumer applications are later. Microsoft, Intel, Google, Apple, etc. are early entrants to the enterprise, in 2009 E3 Game Show, Microsoft with xbox 360 host released the first generation of 3D visual perception products based on structured light technology Kinect depth camera, for capturing the movement of the human body in three-dimensional space, to achieve human-computer interaction through posture. Intel released RealSense, a structured light technology-based product, in 2013 for gesture recognition, facial analysis, background removal, and 3D scanning.
In 2013, Apple acquired PrimeSense, becoming the third company to mass-produce consumer-grade 3D structured light depth sensors after Microsoft and Intel, and in 2017, it was installed on the iPhone X and subsequent series of large-scale volumes, which promoted the development of the 3D structured light industry.
Startups in the field of domestic 3D vision have actually made progress before Apple released the iPhone X. For example, Obi Zhongguang, founded in 2013, has completed the mass production of Astra, Astra Pro, Astra mini and other series of 3D depth cameras as early as 2015, and in 2016, it has obtained a strategic stake from MediaTek, becoming the fourth manufacturer in the world that can mass-produce consumer-grade 3D structured light depth camera sensors.
In 2018, three-dimensional face unlocking has almost become the standard of many high-end smart phones in China, and each company matches its own different AI dot matrix model algorithms, fault-tolerant algorithms, self-calibration algorithms, etc. through relevant hardware to improve the recognition accuracy as much as possible.
For example, the Xiaomi Mi 8 transparent exploration version released in May 2018 and claimed to be the first domestic mobile phone equipped with a front-facing structured light lens adopts the solution provided by the start-up company Mantis Huishi.
OPPO FIND X, released in June 2018, uses the Obi Medium Light solution in its 3D structured light hardware, and the algorithm uses Megvii FACE++, which builds a 3D facial model by projecting 15,000 scattered spots, claiming to be the first 3D structured light technology in the Android camp to approach the terminal of Apple XS.
The Huawei MATE20 Pro released in October 2018 uses a self-developed software and hardware solution, adopts the same VCSEL optical principle as Apple, has the same number of scattered projections (30,000), and declares that the misidentification rate is not higher than one in a million, and can cooperate with multi-dimensional user use scenarios such as 3D modeling.
In the 5 years from the establishment of Obi Zhongguang in 2013 to the large number of 3D visual solutions used in domestic flagship mobile phones in 2018, according to statistics, more than 30 domestic enterprises in the field of 3D vision have started businesses, such as Mantis Huishi, Guangjian Technology, Lingming Photon, etc. The deep pupil of Glining, which recently landed on the science and technology innovation board, also focuses on three-dimensional computer vision technology.
According to PwC data, the global 3D visual perception market size in 2019 is 5 billion US dollars, and the market size will develop rapidly, and it is expected to reach 15 billion US dollars in 2025, with a compound growth rate of about 20% in 2019-2025.
Interestingly, after the 3D structured light boom in 2018, domestic flagship mobile phones in the past two years, such as Xiaomi 11Ultra, Xiaomi 12Pro, vivo X70 series, OPPO Find X5 series, etc., have not adopted 3D structured light face recognition technology.
3D vision technology is still continuously cultivated in the industrial field. In the consumer field, after Apple brought a short boom, 3D vision still seems to be absent from the large-scale application of consumer products, and is only used in a few fields such as sweeping robots.
The second wave of 3D vision entrepreneurship: aiming at automotive and industrial
From 2013 to the release of the iPhone X, it can be called the first wave of 3D visual entrepreneurship in China. Since 2021, whether it is looking at financing or landing scenes, the second wave of 3D visual entrepreneurship is happening.
According to the incomplete statistics of Yiou, in the first quarter of 2022 alone, there were more than 10 financings in the field of 3D vision. In the first half of 2021, there were only 14 financing incidents in the 3D vision field, and the frequency and density of financing in the 3D vision field increased significantly year-on-year.
In 2021, there are also many 3D vision companies such as Toydge Technology, Baochain Intelligence, Xenad, Eva Technology, Seebit Robot, Sea Core Diagram, Lingxi Robot, etc. to obtain financing, and Mercamander Robot even obtained nearly 1 billion yuan of C round financing.
It can also be simply seen from the name of the financing company that the consumer field is not the main front of 3D vision technology. However, in more industrial, industrial, Internet of Things and other fields, 3D vision technology has been commercialized.
For example, Obi Zhongguang released the first standard camera Femto in the iToF technology ecosystem in December 2021, which is designed for the diversified development needs of the AIoT field. Released simultaneously, there is also Alby Zhongguang's first algorithm SDK Orbbec Pose, which can be adapted to the Femto camera to complete a series of 3D application development based on human skeleton algorithms, and has been officially launched in the Obi Zhongguang 3D vision developer community.
It is understood that Obi Zhongguang has landed a variety of 3D perception technologies in consumer electronics, mobile payment, intelligent passage, industrial manufacturing, smart home, intelligent assisted driving and other fields, including innovative application products such as intelligent robots, 3D camera mobile phones, 3D face brush payment terminals, 3D face door locks, etc., and many products have reached million-level mass production.
However, million-level mass production is still not large-scale in the field of consumer goods, and mobile phone shipments are often tens of millions of scale. For example, the number of lasers used in the production cycle of the iPhoneX released in 2017 alone, according to Zhu Li, the founder of Guangjian Technology, exceeded 40 billion, more than the total number of lasers produced by humans before.
Although the shipment scale of mobile phones is tempting, 3D vision still needs to expand more scenes. Many 3D vision companies are also extending the landing scene from mobile phones and face brush payments to more areas such as intelligent security, intelligent manufacturing, and intelligent hardware.
Security is still the main field of 2D vision technology, but with the increasing number of cameras and higher resolution, the bandwidth and computing power required for data transmission and analysis are also getting higher and higher.
Moreover, the projection phenomenon in 2D visual imaging makes the object geometry and surface reflection characteristics, the spatial relationship between the light source and the object and the camera, etc. are synthesized into a single image gray value, and it is difficult to distinguish its stereoscopic information. 3D vision, on the other hand, describes the shape and spatial organization of 3D objects in an object-centered coordinate system, which can convey spatial information that 2D vision cannot distinguish.
Compared with 2D vision, 3D vision actually requires a reduced amount of data, which can reduce the data transmission and energy consumption of traditional 2D camera solutions and save a lot of operating costs. This is a counterintuitive feature – the storage space required by 3D vision to convey richer visual information is not necessarily larger than that of 2D visual information.
3D vision has moved from the industrial field to more fields
In the view of many people in the industry, 3D vision technology will be applied to more industries in the future, combined to produce more high-tech products that change people's lives.
Taking the automobile manufacturing industry as an example, automotive power components are often irregular in shape, heavy in weight, and inconvenient for human operation, which requires the use of automated robotic arms. However, the robot arm faces irregular parts, and it is more difficult to program the action and make fixtures.
For the automation pain points such as loading, unloading, handling, and installation of automotive power components, the combination of "3D vision sensor + mechanical arm" provided by Elson, a provider of robotic 3D vision solutions, is a feasible solution.
From automobile manufacturing to electric vehicle battery manufacturing, semiconductor electronics, food production, industrial testing, as well as warehousing and handling, mixed depalletizing, logistics sorting and other fields of production and manufacturing automation, quality inspection automation, 3D vision sensors have begun to popularize, can achieve face recognition, gesture recognition, human skeleton recognition, three-dimensional measurement, environmental perception, obstacle avoidance, following, three-dimensional map reconstruction, defect detection and other functions.
For example, 3D matching defect detection is to compare the 3D point cloud data of the object collected by the 3D sensor with the standard CAD, and generate a difference report containing height, width and volume data according to the pre-set difference threshold to find the defective product.
Is the biggest demand for 3D vision in the industrial field or the consumer field, in production or logistics?
In fact, regardless of the field, all industries have a greater demand for 3D vision, and related companies are also actively expanding the application scenarios of 3D vision.
For example, Mantis Huishi, which is committed to the research and development of the underlying layer of 3D vision and application development, on the one hand, aims at consumer fields such as smart eyes and smart phones, and develops a new generation of 3D sensors and 3D imaging devices with professional-grade imaging quality and consumer-grade cost, which can be used by AR/VR UGC equipment, or used for 3D face recognition, gait recognition and other scenarios.
On the other hand, it is also involved in logistics, AR smart tourism, smart animal husbandry, 3D medical beauty, automobile collision detection, 3D studio, holographic real-time live broadcast, holographic real-time conference and other services and provide solutions, spanning multiple fields.
For example, the use of three-dimensional space scanning equipment for indoor space digitization can not only provide accurate space size for customized furniture in the home improvement industry, but also provide more intuitive housing information for intermediary industries such as shells.
Smart door locks, building access control and other fields are also the application scenarios of 3D vision.
It is understood that Xiaomi's first 3D structured light smart door lock released in October 2021, Eva Technology may be one of the providers of 3D face recognition smart door lock solutions. Based on the self-developed 3D vision AI special chip, Eva Technology can actually provide one-stop services and modular solutions for more fields: security, sweeping robots, new retail, 3D interaction and other fields.
The market size of security, according to the China Commercial Industry Research Institute, is expected to reach a market size of 1,013.4 billion yuan in 2022.
Sweeping robots need to have the ability to perceive the entire indoor 3D space, formulate intelligent obstacle avoidance strategies, and have the capabilities of spatial modeling, positioning, path optimization, etc., and the demand for 3D sensors is very large. It is understood that the sales of mainland sweeping robots in 2020 will be 6.54 million units, and the expected sales volume in 2021 will be 8.85 million units, which is close to the order of ten million.
Smart door locks have also had a market size of several millions, according to the total data pushed by Aowei Cloud Network (AVC), the sales volume of China's smart door lock market in 2021 is 4.58 million sets. However, fingerprint recognition and even password locks still occupy a high proportion, and 3D sensors need further penetration.
These products with high shipments and demand for 3D vision solutions, such as mobile phones, security cameras, self-driving cars and more industrial products, provide a broad space for the development of 3D vision sensors. At present, the use of 3D vision technology in various industries is still in a very early stage. Most of the 3D vision companies are still in the stage from 0 to 1.
A few years ago, because SenseTime, Megvii, Yuncong, and YITU occupied more than 50% of the market share of the computer vision track, they were called the "Four Little Dragons of Computer Vision". The "Four Little Dragons" have also locked in more people's attention and become a corporate case often seen in the news in the computer vision track.
In the field of 3D vision, which four companies can occupy more than 50% of the market share and qualify as the "four little dragons" in the field of 3D vision?
In the stage of the industry from 0 to 1, the market pattern is not yet stable, and the "four little dragons" may temporarily "hide in the wild". But in the industry explosion stage from 1 to 10, it does not take a few years, or even a year, for the "four little dragons" of the 3D visual track to begin to appear in the rankings.