In the field of VR, spatial positioning and tracking is a key part of the user experience, which can not only provide better immersion, but also significantly reduce the user displacement of the screen out of sync and vertigo. Among the VR headsets that have been launched, HTC Vive's product experience is in the lead, and one important reason is that it has the Lighthouse spatial positioning system, which is based on the "laser light tower + photosensitive dipole array" technology solution, enabling high-precision, high-response speed indoor positioning, so that in the interaction The experience is better than the competition.
Camsense hopes to provide low-cost, high-performance spatial positioning and location tracking systems for industrial, virtual reality, smart TV and other industries to solve the problem of human-computer interaction and high-precision positioning.
In the VR field, Camsense chose to develop a set of Outside-in Tracking spatial positioning solution (Inside-out solution will also be launched soon), through the external infrared camera and feature point tracking, to locate the position of the human head and controller.
Specifically, Camsense will be placed on the VR headset and mobile handle LED active light source, these LED light sources will emit their own light, and different light sources emit different colors of light. Because the infrared camera itself has blocked out most of the visible light information, so that the camera can make a good distinction between the light source and the background environment, the relative position of different light sources when shooting, and the deep learning algorithm then calculates the coordinates of the LED light source in the picture to project the position of the light source relative to the camera, and finally determine the position of the handle and the headset.
Considering that there is a gap between the monocular camera and the binocular solution in terms of positioning accuracy, which is more challenging for the algorithm, Camsense has optimized four aspects.
One is to improve the calibration accuracy by continuous spatial sampling at high frame rates, which can be interpreted as repeatedly taking pictures of an otherwise coarse-detailed image to maximize the number of samples collected and thus supplement more pixel information.
Second, through algorithm optimization, the image extraction accuracy is accurate to 0.01 pixels, i.e. sub-sub-pixel level, which is an order of magnitude (about 10 times) higher than other companies in the industry. The technical logic is to divide each pixel into smaller units and implement interpolation algorithms for these smaller units so that a relatively low pixel monocular camera can be used to achieve a performance that can only be achieved with high-priced hardware, thus reducing hardware costs.
Third, the use of multi-sensor fusion technology to improve the signal-to-noise ratio by incorporating information collected by the IMU inertial measurement unit.
Fourth, by modulating light to remove the effect of distracting factors such as ambient lighting on the camera.
Christopher told Xtecher that Camsense is the second company in the world to use monocular to achieve VR 3D spatial positioning system, and has compared with Oculus Rift which uses the same kind of solution, and has outperformed Oculus solution in terms of average displacement error, average rotation angle error and average displacement jitter, average rotation angle jitter and other indicators, and Hardware components cost is lower. This solution has attracted the attention of HTC, and Wang Xuehong of HTC has visited Camsense this year for inspection.
As for commercial customers, Camsense has signed a cooperation agreement with a domestic VR benchmark customer to provide spatial positioning products; a famous domestic VR benchmark customer is being evaluated, and Camsense ranks first in the technical indexes of the projects involved in the evaluation. In addition, Camsense also signed a product cooperation agreement with a foreign VR customer.
Based on its own system, Camsense has launched two high-precision cameras - Camsense M1, a binocular camera for TV control, TV games, and robot indoor navigation, and Camsense M2, a monocular camera for VR spatial positioning and industrial field positioning, both of which have been shipped in mass production. In addition, the company can also provide cameras and mobile handles for VR games, as well as cameras and game light guns adapted to TV games, produced on a foundry basis.
Smart TV as Camsense's earliest entry into the field, only the game light gun, the company has reached cooperation with Skyworth, LeTV, cool open, Hisense and other mainstream TV manufacturers, so far the cumulative shipments reached more than 100,000 units, product cost pricing is only about a hundred yuan, profit points include hardware sales and SDK charges. The cooperation model is B2B2C - Camsense provides camera and gamepad hardware to TV manufacturers, who then push them to C-end users in the form of matching sales or gifts. This approach reduces operational costs and Camsense is able to focus on the software level of development.
In addition to TV and VR, to take full advantage of its positioning technology, Camsense is working with Shanghai University to develop a solution for high-precision 3D measurement of robotic arms for the industrial sector to minimize robotic arm operation errors, which can be used in flexible automated assembly lines for large precision parts. Potential customers include high-end manufacturing companies such as Commercial Aircraft Corporation of China. The system requires a positioning accuracy of 0.2mm in a large measurement space, a challenge that Camsense's positioning system is up to.
Talking about the future planning, Christopher said that the company will focus on the industrial field and VR device research and development in the next step. Smart TV, as the company's current main revenue source, has established a certain degree of market reputation and will continue to focus on promotion.
CEOChristopher, a serial entrepreneur, master tutor of Shenzhen Graduate School of Tsinghua University, Shenzhen high-level talents, Nanshan District pilot talent, under the supervision of 973 chief scientist, Changjiang scholar Professor Dai Qionghai, research in the field of human-computer interaction for more than ten years, with more than ten invention patents, has worked in Chicago Bell Labs, China Mobile, etc. The chief scientist, Qionghai Dai, is a professor, doctoral supervisor and deputy head of the Department of Automation at Tsinghua University, the winner of the 2012 National Technical Invention First Prize, the chief scientist of 973, the Changjiang Scholar, and the recipient of the Distinguished Youth Fund.
Camsense was founded in 2014, received 5 million angel financing from a famous domestic Internet angel investor in June of the same year, completed tens of millions of Pre-A rounds of financing from Orient Fuhai and LeTV in June 2016, and received 1 million RMB free funding from Shenzhen in October 2016, and has now started a new round of financing, which will be used for technology R&D, team building and market expansion.
According to the latest news, Camsense has successfully passed the selection of Tencent AI Accelerator in the ratio of nearly 1000:20 and will soon be enrolled in Tencent AI Accelerator, which will provide five major levels of entrepreneurial support including AI technology, capital, mentorship, channels and marketing.
