With the increasing proportion of video resources and content in the Internet, the corresponding contradiction between the utilization level of video resources and the increasing business and product demand will become more and more prominent. This article is from the speech shared online by Qi Yankun, product manager of Huawei Cloud video access service. Combined with practical experience in video access service, we discuss the trend of cloud on video, pain points and the exploration and practice of Huawei cloud video access service.

By Qi Yankun

Organizing/LiveVideoStack

Hello, my name is Qi Yankun, from Huawei Cloud Media Service Product Department. I am currently the product manager of Huawei Cloud Video Access Service. Today’s content to share with you is divided into the following parts.

The first part is the development trend of security video, including technology trend, national policy, etc. The second part is the pain points we found in practice; The third part is to share the cloud application practice; The fourth part focuses on the scene and some practical cases; The fifth part has a small video demonstrating the operation process on the cloud; The sixth part is communication and interaction.

1. Security video development trend

Cloud and AI have opened a new era in the security industry.

Security is closely related to everyone’s life. There are camera related systems in the living community and the office building. These are the most basic security. From the perspective of industry, security has experienced about half a century, and the industry has developed for a long time. From the analog signal at the very beginning, the problem of “visible” was solved, namely “something/nothing”, and the “thousand-mile eye, wind-following ear” in the Chinese myth of Journey to the West was realized. After that, the whole technology trend and business development still fail to meet customer demands. For example, a company has several subsidiaries in different regions, and the boss wants to view the subsidiaries in different regions remotely. This is the Internet era. The combination of the Internet era and the mobile Internet has greatly improved the security experience, but it has also brought a big problem – bandwidth, to transfer video to the cloud to solve the bandwidth problem, this is the key technology.

Then people entered the era of intelligence, artificial intelligence is also closely related to our life, such as parking lots do not need to pay fees, only need to recognize the license plate.

Why cloud and AI have opened a new era? First of all, AI reduces the need for human resources and improves efficiency. If washing machines enter thousands of households and free people’s hands, THEN AI can free people’s eyes in the field of security. The other is the cloud level, no cloud can of course, but the construction of resources in the later stage including more and more video, from STANDARD definition to HD, from 2D to 3D and interactive, this piece of the network, computing power demand is more and more, so “cloud” is a good trend.

I just mentioned that we are entering the smart age in security, so what role will video play in the smart age?

The intelligent world is divided into three layers: perceptual, interconnected, and intelligent. From the perspective of perception layer, it is mainly about data acquisition. We have obtained data from some analysis agencies that 65% of the data in the digital transformation of the industry is video data, enough to see that video is a very important part of the entire digital transformation of the industry. Another statistic is that in our smart world, 49 percent of the apps are based on video, which can be seen as the core of the perception of everything, the clairvoyant of the smart age.

The cloud brings several benefits: sharing, openness, intelligence, and unified management. Without the cloud, the aforementioned resources are one problem. Another problem is that a lot of hardware is stacked together and data and applications, software and hardware can’t be decoupled. For intelligent evolution, as the amount of video is increasing, more demands are made on the GPU. Therefore, it is difficult to process the video locally. If the devices are located locally, it is difficult to manage the devices in a unified manner because the devices of a company with many subsidiaries are located in different computer rooms. So from now on, the cloud is very good for sharing, openness, intelligence and management.

From the national policy level, whether the Ministry of Public Security, the Ministry of Transportation, the postal industry, the dangerous goods production enterprises, the country has a clear letter, hope that enterprises can achieve video networking and video sharing. After the interpretation of the policy, we summed up the following four levels: the first is the whole network sharing, only after sharing, everyone’s data can be unified, do AI analysis, and do linkage and sharing with other equipment; The second is global coverage, mainly emphasizing the reasonable deployment of video in some regions; The third is the availability of full-time, recently some video website problems of the news frequently hit the headlines and hot search, this is not to do the availability of full-time, if they can be on the cloud, in the availability of the piece will have a good performance; Fourth, the whole process can be controlled. First of all, privacy, data can not be stolen is the most important. In addition, if you want to see the data, you need to obtain authorization from yourself.

Therefore, from the perspective of national policy interpretation, whole-network sharing, whole-area coverage, whole-time availability and whole-process control are important information.

2. Analysis of cloud pain points on security videos

We’ve also seen some problems in recent projects we’ve been involved in, especially when companies buy cameras from multiple manufacturers and ask us how to connect to the cloud because each camera has its own protocols.

Many well-known camera manufacturers in order to better service their products as well, will provide the corresponding video access platform, which can lead to a problem, if from the customer perspective, the multiple camera manufacturers have more than one system, from the enterprise point of view this is relatively fragmented, necessarily hinder for data sharing. The second point is the protocol, the current video protocols are RTMP, RTSP, Onvif, RTP, GB/T28181, etc., will cause trouble to enterprise customers. The third point is that customers will report that the video lags, splashy screen, large delay and frequent flow interruption, which are all problems encountered in actual projects. The fourth point is the cost, can be seen from two levels, first, if deployed in the local, then the demand for computing power network is relatively large; The second is on the cloud, based on the cloud to build BGP cost is high.

I would like to share with you some of my thoughts on these challenges.

First look at the agreement, the figure on the left you can see from the network layer (IP) to the transport layer (TCP, UDP), in the above there will be a lot of corresponding to audio and video transmission protocol, you can see from the Angle of audio and video transmission is divided into three levels, and at least the first point is signaling control, including h. 323, this is in agreement with more traditional telecom industry, In the Internet layer is used more SIP protocol, in order to the whole audio and video transmission effect is better, there will be RSVP, RTCP and other protocols, there is a transport layer protocol RTP, this is a relatively low-level protocol.

There are many upper-layer protocols, such as HTTP, RTMP, HLS, GB/T 28181, etc. What do these protocols do? First of all, GB/T 28181 is an agreement made by the Ministry of Public Security of China (which will be highlighted below). Onvif and PSIA are also based on security systems, which are widely used in overseas markets.

In addition to a few more typical protocols, RTSP latency is better, but it is more complex, and Web support is not very good, from these characteristics can be seen that it is more suitable for security protocol. RTMP divides the data into many small pieces for transmission in the process of data flow, which will lead to a lot of delay, about <3s, but it has good support for Web, you can view it on the Web through plug-ins, which is more suitable for live broadcast scenarios. HLS runs well from a cross-platform point of view, but with high latency.

To sum up, we can see that different protocols have different positioning, so in the security system or in the business development, how to choose the right protocol to meet the demands of business development, then there is the following GB/T 28181 protocol.

The protocol itself was formulated by the Ministry of Public Security of China. Currently, there are two versions, GB/T 28181-2016 and GB/T 28181-2011. The main purpose of this protocol is to solve the problem of interconnection between camera systems of different manufacturers and regulate several important scenarios. From this point of view, GB/T 28181 is a very suitable agreement to solve the problem of docking between different manufacturers. GB/T 28181 actually defines SIP, SDP and RTSP, which can be considered as a protocol specification. The protocol mainly defines 12 scenarios. The device can register and log out with the server, play real-time audio and video on demand or playback of historical video, control the device (camera rotation direction), and notify the location of the device status change.

With the GB/T 28181 protocol, from the perspective of Huawei Cloud, we give priority to the implementation of this protocol. All the mainstream cameras in the current project can be connected based on GB/T 28181, which can well solve the problem of multi-vendor and multi-protocol. We can also support standard protocols such as RTMP and HTTP-FLV if different business scenarios require special requirements.

GB/T 28181 is also very simple from an on-device cloud perspective, requiring only a few simple configurations and no code development to connect to the cloud. After the video is connected to the cloud through the protocol, we also provide some capabilities on the cloud, including device status management (online and offline), video stream management (video stream activation and deactivation), automatic invitation of abnormal video stream, cloud storage, watermarking, screenshots, transcoding and other capabilities combined with other cloud services. The most important is AI blessing (more on that later).

From a video point of view, the experience is a very important thing. If video broadcast caton flowers screen is hard to accept, from experience, first ruled out end side issues, to ensure that local video playback smooth, after excluding the local problem, the key to check the network reason, now, it seems, from video access, can adopt different when connected to the network, using network access video is stable, the home broadband effect is poorer. From the cloud point of view, the bandwidth reservation is very sufficient.

Then there are outages, for a number of reasons, possibly because the system itself is abnormal. Huawei cloud’s solution is to use the automatic invitation capability, the system automatically identifies which streams are broken, and then automatically pulls them up, and the customer experience will be greatly improved. Some disconnections are required by customers based on service characteristics. In this case, the solution is for the cloud system to automatically identify the disconnections and notify the customers through active subscription. The upper-layer application will handle the subsequent actions.

In fact, the configuration of GB/T 28181 is relatively simple. The camera itself is a small system, which requires login and some configuration, which may be complicated for the construction personnel. The solution is to scan the code for registration. Device information is managed on the end based on qr code scanning and cloud registration is completed, which intuitively reduces the configuration complexity. If you are familiar with GB/T 28181, configure it on the device.

Cost is of great concern to enterprises, and the cost of cloud on video is roughly divided into two parts.

The first part is the enterprise bandwidth cost of enterprise customers, which needs to be negotiated with the operator. The main method to reduce the cost on the access side is to reduce the transmission bit rate. The second part is the cloud bandwidth cost, bandwidth costs to talk about the clouds, it must be mentioned huawei greetings native media network, we run in the cloud native media network a lot of business, including live, on-demand, real-time audio and video, here there are both big cloud edge processing, it is formed by large clouds and edge of the network media. Based on this kind of media network, we can reuse network bandwidth resources very well.

From the perspective of cloud video access, it is divided into several parts. Based on cloud unified scheduling, signaling messages are first processed by the large cloud, which is the signaling scheduling process. If the access is legitimate, the camera or NVR will obtain the access point from the cloud. This access point may be an address on a different edge and is scheduled based on a proximity policy or a reliable signaling policy. In this case, the cameras that are originally directly dispatched to the large cloud for BGP access will preferentially go to the edge, and then the edge will pass the special line to send the video to the large cloud for corresponding EI analysis or storage, which is the overall architecture. From a reliability point of view, if the point of edge access fails, the video is dispatched to other convenient nodes that are providing reliable service. If there is an extreme case where all the edge nodes fail, then we will dispatch the video directly to the cloud. So from the perspective of the whole architecture, it can guarantee the access reliability and also guarantee the lowest cost.

3. Huawei’s cloud application practices for video access

I have just analyzed the three problems we encountered in the delivery process of the project. So what else can video do when it’s plugged into the cloud? In Huawei Cloud, especially the services combined with video include live broadcast acceleration. If the video is connected and wants to be watched by many people at the same time, it can be combined with live broadcast acceleration service. The success rate of Huawei live broadcast is >99.7%, the lag rate is <2.5%, and the playback delay is <3s. In the cloud, Huawei cloud has a good competitive advantage with its storage capacity and service availability of 99.995%. There is also a combination point. If you want to do some on-demand transcoding or hd/low code conversion after the video is uploaded to the cloud, we also have supporting video services. The other is the combination of intelligent analysis services, which I will focus on next.

Here I listed some intelligent algorithms related to daily life, divided into seven categories: face recognition, the gate of the office building to use this service; Intrusion detection, cell use more, if found intrusion can be identified and alarm; Personnel statistics, whether the personnel gather, the urban management scene can be applied; Safety detection, applied to fire and smoke alarm; Vehicle detection, illegal parking detection, traffic flow statistics, etc.; Human identification, human distribution control, abnormal behavior and so on are usually applied in urban management.

You can use Huawei Cloud services in different service scenarios.

In view of the above content, I would like to make a summary based on the three pain points we encountered in the early stage: multiple manufacturers and multiple agreements, experience and cost.

The first is cloud enablement, which combines cloud, AI and storage and codec. The second is full real-time, multi-edge media network to ensure business continuity, if there is an exception, can do automatic switching, to provide automatic invitation. Third, wide coverage, compatible with multiple manufacturers, in some projects, including Hikon Dahua, Huawei and other mainstream cameras and NVR equipment have done docking. The fourth is low cost, high cost performance, we combined with the media service network can achieve low cost. The fifth is the full scenario, from the point of view of the support scenario, we are the full scenario access, including security, traffic, park, business and super can support.

Let’s discuss the security scenario.

4. Scenario and case introduction

From the perspective of video access services, the main scenarios include smart stores, home security, smart parks, smart environmental protection, and traffic networking.

Intelligent stores are generally remote patrol, customer flow analysis; Home security, now seems to be a trend, the installation of cameras in the home to observe the situation; Smart park, widely used, park visitor management, license plate recognition, etc. Smart environmental protection is also an increasingly important scene. The following will focus on the case of solid waste monitoring; Traffic networking is also an important scenario.

Here are some examples of cases that have been delivered.

Bright kitchen stove and everyone’s life is closely related. Now the parents are very concerned about the students in the school dining, this is a big appeal, the second is the delivery is more and more popular, everyone wants to see food processing process, if these are arrange manpower regulation will form a lot of labor costs, cannot reach such a large scale, so video to the cloud and big data analysis, through the analysis of the AI It can automatically identify abnormal violation scenarios and present them to relevant supervision and management departments.

In the scene of bright kitchen stove, there are nine main intelligent identification scenes: no mask, no chef’s hat, playing with mobile phone, smoking monitoring, shirtless, harmful animals, camera cover, uncovered trash can, and no gloves. The combination of video, AI and big data can greatly improve the efficiency of regulation.

Wisdom venues and related to children, there are more and more training venues, parents will be recorded in the process of accompanying video, this process there is a large amount of video recording, reading trouble, through gathering venues video cameras, and then allowed entry like children, under the premise of based on human to search the related brilliant moment, through the clips make relevant video pushed to parents, Parents receive more wonderful pictures, easy to find and easy to store.

Once solid materials produced by chemical enterprises are stored and transported, improper disposal will lead to major problems. The most intuitive yancheng explosion was caused by improper disposal, and the local video was also lost. From this point of view, the video cloud can eliminate the problem of video loss. In this business scenario, not only video applications, but also other IoT devices are associated with the perception of solid waste lifecycle management, which is a good way to monitor the enterprise and help standardize production.

These are some typical cases.

5. Operation demonstration of video access to the cloud

Here’s a short video on how to quickly video onto the cloud.

(Video quick cloud operation demonstration)

As you can see from the video, the cloud is very simple, with no code to write and just a few steps to configure.

6. Summary

Let me just give you a quick summary of what I’ve been sharing today.

The first is that security has entered a new era, into the era of intelligence.

The second is the benefits of the cloud on video: sharing, development, intelligence, unified management.

Third, Huawei cloud video access service itself includes support for national standard, RTMP, HTTP-FLV and other protocols, easily on the cloud.

Fourth, after the cloud can be combined with cloud services, especially algorithms, storage, live streaming services, on-demand services, etc.

That’s all for today, thanks!

Q&A:

1. Is the GB/T 28181 scheme your main access scheme?

GB/T 28181 was developed by the Ministry of Public Security of China and is widely used in domestic video management. More access protocols used overseas are Onvif, PSIA and other protocols. Therefore, GB/T 28181 focuses on the domestic market, which is to solve the problem of management cost caused by different manufacturers and different agreements.

2, for some stock of cameras, feel compatible is not very in place?

This is a real problem, and we now find that the camera itself has support for the GB/T 28181, some of which are older than the 2011 version. There are two main versions of GB/T 28181. The first is the 2011 version, which is generally supported by older cameras. The new camera is supported by the 2016 version, and the protocol of the 2016 version will be more perfect.

3. Will SRT, QUIC and other protocols be supported later?

The video access service of Huawei Cloud will always develop in the future. We will combine the business characteristics and customer needs. Any technological development will be combined with these two points.

4, just said access protocol support GB/T 28181, like some small device vendors or previous old models do not support how to deal with?

If you really do not support GB/T 28181, then you can find an NVR, because now NVR can support, through a small device to access the NVR first, then access GB/T 28181, because NVR support GB/T 28181, equivalent to a conversion. If you don’t have an NVR that supports GB/T 28181, you can do a conversion at the local gateway and the cost will be slightly higher.

5. Are the projects today delivered in public cloud or private cloud?

Today’s examples are based on public cloud delivery. Today, the cloud I mentioned here includes public and private clouds. The action of “cloud on” does not necessarily require the public cloud, but private cloud can also be used.

6. Can STREAM H.265 be accessed?

You can.

7. Can the monitoring stream be encrypted, encrypted after the stream media is received, and decrypted at the player end?

The GB/T 28181 agreement needs to be improved. State secret algorithm can be used for encryption and decryption.

8. What if the camera only supports UDP push streaming?

GB/T 28181 supports two protocols: UDP and TCP.

9. How about the real-time performance of the video after using the cloud?

From the current project, if the local bandwidth is good enough, it is no problem to play the video within 1-2s on the cloud, and the real-time performance is good. The experience will satisfy most business scenarios.

10. For the remote configuration of parameters on Dahua and Hikon cameras, do we directly call the MANUFACTURER’s SDK?

GB/T 28181 protocol itself does not need to use SDK, you can directly log in the camera for configuration.

11, GB/T 28181 protocol uplink security issues? Heard that the upside is easy to be hijacked?

To achieve better security, you can harden the link layer to ensure data security on physical channels.

12. Does the video access service support mobile real-time streaming? What are the performance requirements for the terminal?

Huawei’s video access service supports three protocols: NATIONAL standard, RTMP, and HTTP-FLV. Streams can be pushed based on RTMP. General mobile phones can meet the demand, most of the live streaming is carried out through mobile phones.

13, the public cloud is similar to mouse detection, these can be detailed? Each frame detection cost is relatively high, frame extraction will miss detection.

This is a good question, my team is more focused on video access services, mouse detection is handled by a different team, and I can take this issue back and discuss it with the EI team.

14. After the videos of H.265 are sent to the cloud, shall H5 directly use H.265 to play or transcode into H.264?

In theory, either could work.

**15, ** just showed 40+ kinds of algorithms are All made by Huawei? Or was it developed in cooperation with ecology?

They are mainly made by Huawei itself. There are many combinations of EI algorithms of Huawei Cloud and partners, and they will also introduce partners’ algorithms.

16. Can you achieve voice talkback?

The GB/T 28181 itself supports voice.

17. I think some companies are using RTC, which is more real-time and more compatible with the network. Does the teacher recommend this way? The service requires the video to be on in seconds.

In some scenarios, such as smart doorbell, RTC solution is recommended because of its low latency. If the delay requirement is not high, there is no need to use the RTC solution, and the solution is selected based on the service scenario.

18. What is the maximum number of streams that can be used concurrently with an existing project?

Huawei Cloud single tenant can concurrently log in to 100,000 levels.

For details, please scan the __ qr code __ or click __ to read the text __ for more information about the conference.

This article uses the Article Synchronization Assistant to synchronize