Behind the hot live broadcast of the world cup, the participation of cloud computing manufacturers can be said to be very high. All cloud computing manufacturers provide hardware devices or software algorithms to varying degrees. The Tencent cloud AI audio and video technology behind the CCTV World Cup high-speed HD live broadcast is not well-known, but few fans know that all video streams are distributed after being processed by Tencent cloud.
“For video enterprises, bandwidth expenditure accounts for a large part of operating costs. If you want to provide users with high-definition video experience, you must increase the bit rate of video stream, and the corresponding total cost of bandwidth will be further increased. And high speed HD is solving this problem, and can provide users with more HD video at a lower bit rate. ” Tencent cloud Senior Product Manager Li Haiqi said.
At present, there are some problems in common transcoding, such as transcoding distortion, resolution blur, lens jitter, noise and sawtooth block. Tencent cloud high speed HD technology, also known as “intelligent dynamic coding technology”, mainly uses machine deep learning, through video scene intelligent classification, coding parameter matching, pre-processing, coding dynamic optimization, code rate intelligent control, detail Reduce, ROI processing and other technologies and processes, intelligent processing of live or on-demand video stream to match the optimal coding parameters, so as to provide users with more HD live video stream at lower bandwidth cost.
In the aspect of intelligent classification of video scene, more than ten mainstream categories and dozens of sub category scene model libraries are formed through deep learning. During the high-speed HD technology service, the live stream can be detected and analyzed in real time, and the corresponding scene model can be matched. In terms of matching rate, CNN network models with obvious scenes such as games, football, basketball, animation and so on have a matching accuracy of more than 98%; in scenes with scattered picture features such as TV series, outdoor sports, food, tourism and large inter frame motion changes, CNN + RNN + LSTM can ensure a matching accuracy of more than 85% by analyzing the time domain and spatial domain;
In terms of coding parameter matching, Tencent cloud extreme HD technology will select the optimal coding parameters according to the real-time recognition results of scene classification, combined with video source rate, frame rate, resolution, texture and motion change amplitude, as well as the dimensions of machine load and image quality effect.
In the aspect of pre-processing, according to different scene classification, customers’ different requirements for video quality, as well as the combination of video source image texture and real-time motion detection results, pre-processing such as sharpening, soft blur, anti interleaving, deblocking, noise reduction, color scale compensation and frame reduction are carried out; Tencent cloud extreme HD technology also supports dynamic optimization of coding, which can be used for different video categories and the same video Different encoding parameters are applied to different video segments in the video frame, and the encoding parameters can be updated in real time according to the frame.
In addition, the intelligent rate control of high-speed HD technology can increase the vmaf value of subjective viewing score by 3-5 points Reduce technology can optimize the noise macroblock with very low CPU consumption, while retaining the integrity of the clear macroblock; ROI processing technology, through the linkage of sift + differential image + moving target image detection, can strengthen, sharpen, color level compensation and other processing of the macroblock coding in the POI area.
In order to avoid the bias of cloud vendors, CCTV hopes to control fairness from the source, which requires privatization. Tencent cloud has come up with a new privatization scheme in the case of insufficient time. It has set up a complete set of Tencent cloud high-speed HD transcoding cluster in CCTV network to support cctv world cup live broadcast on CCTV5 channel, RTMP / flv push-pull source stream, and support CCTV RTMP / flv / HLS pull stream after transcoding processing.
The actual effect proves that the streaming effect of Tencent cloud scheme exceeds the expectation, and the broadcasting effect has obvious advantages compared with other platforms. Lei Feng learned that, so far, Tencent video cloud has served more than 80% of the head live platform. Tencent audio and video AI product solutions can effectively save bandwidth resources for online video, pan entertainment live broadcast, short video and other industries, improve viewing experience, help enterprises save costs and ensure business security.