The design concept of EasyCVR, an intelligent basic platform for general video (security video, Internet video, video big data)

After working on video technology and products for so many years, there have been several problems that I have not solved well:

  1. Is there a platform that supports all access video types? Push stream, pull stream, national standard stream, file stream;

Nowadays, streaming media platforms with various protocols are emerging in an endless stream, including platforms that I have made many protocol types. For example, the EasyNVR video edge service is mainly used to access RTSP and Onvif streams, and the EasyDSS live service is used to receive RTMP push streams. As for processing video files, the EasyGBS national standard video service is mainly used to process the equipment access and output of the national standard GB28181 protocol. Each service has its own application scenarios, which will always exist and cannot be discarded;

  1. How to take out the video in the device and play it back? RTSP playback stream? National standard playback? What about the old equipment?

At first, I was obsessed with the standard protocol for streaming. Therefore, when the national standard GB28181 was not so universal, I always wanted to solve the problem of device playback streaming through Onvif's video query and RTSP video playback. However, Onvif's playback protocol was too It’s not universal anymore, and it hasn’t been popularized until now. So, I later chose the national standard GB28181 as the choice of playback protocol. Later, two problems were discovered. One is that the national standard is not universal, and there will still be old devices that don’t support it. After all, GB28181 only started in 12 or 13 years, and it took a long time to gradually take shape. Second, the playback effect of the national standard is not very good. For video playback with high data accuracy requirements, especially when video analysis is required frame by frame , The playback of the national standard GB28181 protocol is not a good choice. In the end, I succumbed to the SDK mode of the manufacturer~~~

  1. Those who can stream media don't know how to use AI, and those who can use AI may not be good at streaming.

It can be said that video AI is very hot and requires a lot of manpower, technology and financial support to do it well. However, audio and video streaming requires a lot of time and manpower to develop and stabilize. AI vendors want to quickly land in the vertical field, but Algorithms are not enough, and a good video platform system must be supported. In this case, the market advantage will be transferred to the established video manufacturers such as Haikang and Dahua. They have both industry experience and video platform technology. With strong financial support, in order for many AI vendors to better and faster implement AI smart applications, I will make a video platform service that can be easily grafted to AI;

  1. The entry barrier for intelligent video analysis is too high. Even with the BAT SDK, it is difficult to produce results quickly. Is there a ready-made solution that can encapsulate the AI ​​algorithm in accordance with the regulations and throw it on the platform to become a video analysis plug-in?

Most of the development, especially the development of each team, hope to be able to do their own system with peace of mind and avoid too much coupling with other systems. Many AI algorithm customers before us are streaming our videos. Platform services are called in PaaS mode, and AI intelligent analysis is grafted into the SaaS layer. There is no problem in terms of feasibility. However, in terms of architecture, the output of AI intelligent analysis should also be at the PaaS layer. AI intelligent analysis involves a lot of The operation of audio and video encoding and decoding, graphics and images is another skill that is difficult to possess in SaaS application development. Invisibly, PaaS and SaaS are glued together. If there is a video PaaS platform service that can graft AI intelligent algorithms, it is like Like the module plug-ins of the PaaS platform, the AI ​​intelligent analysis and the callbacks of the analysis results are built in the plug-ins, and then the callbacks are received and processed through the SaaS layer. Isn't it beautiful?

Yes, the video analysis plug-in in the video platform! This is what we are going to do;

After analyzing the problem, we came up with a solution. If we can build a video platform where the algorithm part of video analysis can be used as a plug-in of the platform and grafted to the platform, can we use whatever algorithm we want? Algorithm, remember the EasyAIFilter we used in EasyNVR before?

What is EasyAIFilter? https://blog.csdn.net/EasyNVR/article/details/101427700

Yes, that's it. Through the parameter setting of EasyAIFilter, we let the AI ​​algorithm development students encapsulate what they want to do, and finish it in EasyAIFilter. EasyAIFilter can analyze the live stream (full-time, time-sharing, timing), or Analyze the video stream (select the time period). In this way, EasyCVR makes a fixed AI management framework, and the specific AI implementation and AI stress actions can be implemented in EasyAIFilter;

Make an analysis plan, do regular and quantitative analysis of live streaming and real-time streaming. The analysis rules are done by selecting intelligent analysis plug-ins, which can be produced by the manufacturer;

Guess you like

Origin blog.csdn.net/xiejiashu/article/details/109316667