AI has swept all industries. As the two main entrances of the AI era, the intelligent voice industry has entered an explosive period. Driven by new infrastructure, computer vision is becoming the largest and most widely used field of AI.
In the field of voice and text intelligence, whether it is Qimo cloud customer service for the public cloud market or RongCC for large and medium-sized institutions, Ronglian has been leading industry innovation and building a closed business loop based on smart customer service and smart contact centers.
In the visual field, how does Ronglian extend and commercialize its original capabilities? Recently, CPS Zhongan.com interviewed Chen Kaijie, Director of Ronglian Cloud Communication AI Vision Solutions, and exchanged and discussed topics such as communication services, AI technology, and intelligent video surveillance. The following is the content of the interview:
Q: What are the core technologies and product advantages of Ronglian Cloud Communication? What is the difference between the visual intelligence service proposed by Ronglian and traditional video surveillance?
Chen Kaiyao: When the application of voice, text, and video intelligence in communication services has been widely used, Ronglian realized that the current market has a large demand for image recognition, so Ronglian started the research and development of computer vision (CV) .
Since CV and NLP deep learning are technically compatible, and Ronglian has accumulated NLP engine related technologies, it quickly mastered the related technologies of CV scenarios.
Ronglian’s advantage lies in the enhanced processing of data in specific scenarios, the combination of AI and traditional technologies, and the ability to engineer cutting-edge technologies.
For example, the accuracy rate. Taking smart construction sites as an example, Ronglian’s "Smart Eye" visual analysis platform can perform all-round inspections in combination with the complex site of the construction site, and the accuracy rate of Ronglian products can exceed 90% when testing hard hats.
AI algorithms are extremely dependent on data, but effective data collection is difficult in many cases. Ronglian's own data enhancement processing technology can achieve better results when there are fewer materials.
Deep learning technology relies heavily on training data, and the output results of deep learning generally have "threshold" restrictions. In practical applications, it is difficult to set a uniform threshold for all scenarios.
At present, AI technology is still restricted by data volume, data quality, labeling cost, data domain changes and other issues in the process of landing. It is difficult to solve actual CV problems in complex scenarios by relying solely on deep learning models.
Ronglian has a series of engineering optimization methods for actual application scenarios, including image timing analysis mechanism, detection target attribute filtering mechanism, target tracking and ReID mechanism, etc., to improve the accuracy of the algorithm when it is applied.
In addition to technical barriers, an important barrier for Ronglian products is its accumulation in the industry, because video recognition in a certain industry is essentially the accumulation of industry data and the understanding of the industry. In actual operation, accurate insight into the pain points of the industry is a long and important thing.
In addition, regarding the difference between visual intelligence services and traditional video surveillance, the general feature of traditional technologies is that they are highly versatile. The visual intelligence services proposed by Ronglian combined with deep learning can obtain neural network model output based on traditional video surveillance. The result with semantic information improves accuracy while maintaining versatility.
For example, the target tracking scene, using deep learning to detect the position of a specific type of object in the image, combined with traditional background modeling, frame difference method, optical flow method, etc., to achieve a balance and coordination between versatility and specificity.
Q: Which industries are the commercialization of Ronglian CV focused on? What is the focus of these industries, what problems have been solved, and what value has been enhanced?
Chen Kaijiu: Currently, Ronglian focuses on industries such as smart communities, gas stations, chemicals, urban management, health supervision, and bright kitchens. These industries have different focuses.
Take the surveillance scene as an example, 50 cameras, 24 hours a day, produce 1,200 hours of video, which is 36,000 hours a month. Even if the monitoring staff work in three shifts, 100% coverage cannot be achieved. It is easy to miss some emergencies during the period, and it will be difficult to trace the incident after the incident.
Another example is the manufacturing industry production line scenario, where workers have specified operating specifications and specified locations, usually supervised by supervisors. However, the production line is very long and it is constantly moving, and it is difficult for supervisors to cover all the work stations for supervision. Defects caused by improper operation of the product will directly affect user evaluation and have a negative impact on the company.
We use smart vision algorithms to intelligently supervise the key scenes of the above industries, improve management efficiency, coverage, and improve the level of safety management and control, while reducing risks and reducing supervision costs.
Q: In this market, there are hardware vendors, overall solutions, and system integrators focused on end users. What is the difference between Ronglian and upstream and downstream vendors in the industry? What kind of cooperation is there?
Chen Kaiyao: Ronglian adopts the model of AI capability + vertical industry + service, and has accumulated large customers in multiple fields. It is good at digging into user scenarios and discovering pain points; the pain points of a single user may be industry pain points, customized for the pain points Ronglian’s goal is to transform the solution into an industry solution, and to improve the overall industry intelligence level.
In addition, Ronglian’s product design adheres to the principle of loose coupling and supports flexible splitting and reorganization between modules, which can be provided separately or flexibly integrated with upstream and downstream manufacturers’ products to customize solutions according to customer needs.
Q: I understand that Ronglian’s visual algorithm mall is very rich, and there are still scenes and algorithms being explored. What are the undeveloped industry needs that Ronglian will pay attention to in the future? Can you briefly talk about future trends?
Chen Kaijie: There are still some industries that have not yet been developed, such as health supervision, education, land and resources, military industry, port affairs, logistics and so on.
The prediction of future trends is from a technical point of view: a future trend worth paying attention to is from recognition to understanding, and to apply the ancients is from knowing what is happening to knowing why.
In the past ten years or so, computer vision has made significant progress in recognition, but now only recognition is far from what we expect, or just the first step towards intelligence.
The result is still limited to the value of the company to the industry. The result can be explained and the value will be greater. Establishing the connection between objects and objects, objects and environment, with such a relationship from basic attributes to objects to environment, it is possible to realize from knowing what is happening to knowing why.
The most important trend in the future is to achieve the effect of a knowledge map from recognition without knowledge support to understanding that requires knowledge support, or it can be called "computer vision map".
From the perspective of market scale: the current growth rate of the computer vision industry continues to maintain a relatively high growth trend. Deep learning and convolutional neural network technologies are driving computer vision technology, and at the same time driving the rapid development of the entire artificial intelligence industry. Technology has played a pivotal role in all walks of life.
With the improvement of technology maturity and the rapid improvement of hardware performance, classification and segmentation algorithms such as face recognition, object recognition, process recognition, and complex scene recognition are not considered to improve accuracy. In the future, there will be more scenes that can apply computer vision technology. Visual enterprises should explore more vertical field needs and solve their pain points under the premise of strengthening technology.
Q: The security industry is affected by intelligence and information technology, and the boundaries are gradually blurred. Pan-security has become the future trend of the industry. Many giants cross-border entry and industry competition has intensified. How will Ronglian respond to such a fierce market structure? What adjustments will be made to future strategies and plans?
Chen Kaijie: In the era of pan-security, whether it is technology or market, the security industry is no longer a closed industry.
The first is about product upgrades. Ronglian will focus on three aspects to innovate:
First, let security change from passive prevention to active early warning, from "seeing" to "doing"; using visual intelligence to gradually transform the security system from passive recording and inspection to automatic analysis, active early warning or immediate disposal.
Second, make smart security cloud-based. One of the major problems facing the security industry is the “fragmentation” of scenes. This is a problem that cannot be ignored. A large number of fragmented scenes bring about a large amount of data accumulation. The amount of data stored in the security system increases exponentially. At this time, applications will gradually Turning to the cloud for processing and cloudification of complex tasks can not only increase processing efficiency, but also allocate resources reasonably. Ronglian will also conduct in-depth research on application scenarios in the cloud field.
Third, the application scenarios are implemented. The security industry has focused on the functions and performance of various AI algorithms from the previous two years to focus on the matching and connection of algorithms and specific businesses, pursuing the implementation of AI performance to specific business applications and forming specific solutions. Ronglian will devote itself to solving various vertical application scenarios.
At the same time, with the rise of 5G, AI, and policy support, we believe that edge scenarios will bring fresh vitality to the security industry, so the integration of cloud, edge, and end is also an important product plan for us.
It is reported that on November 5, "Ronglian Cloud Communication" announced the completion of US$125 million in Series F financing, led by China's state-owned capital venture capital fund, New Oriental Industrial Fund, Mirae Asset (Future Asset), Blue Teng Capital and CloudAlpha and other strategies Participate with financial investors. Tenda Capital, Citigroup Global Markets Asia Limited and China Merchants Securities acted as financial advisors.
This is the largest private equity financing so far in the domestic cloud communications field.
Ronglian said that after the completion of this round of financing, Ronglian will further increase its investment in technology research and development in the direction of intelligent communication cloud services, increase product thickness and competition barriers; at the same time, continue to innovate and expand the boundaries. In the wave of 5G and new infrastructure, Better help enterprises and government organizations realize digital transformation and intelligent upgrade, and promote the transformation of China's corporate communications market.