Implementation Principles of Real-Time Audio/Video Service

Real-time audio/video uses the WebRTC protocol stack, supporting both peer-to-peer (P2P) and server‑relay (peer‑server‑peer) communication modes.

Problem discovered : When switching from P2P to server‑relay, recorded videos contain rotation metadata, although live playback appears normal.

Problem tracing : The upstream peer’s RTP stream carries rotation information embedded in the SDP; the rotation is negotiated so the receiver can render the video correctly.

SDP extension : Rotation data is stored as an extmap attribute, e.g., a=extmap:4 urn:3gpp:video-orientation , occupying the fourth RTP header extension ID.

RTP packet format : An RTP packet consists of a fixed header and an optional header extension. The X bit indicates the presence of an extension; CC indicates the number of CSRC identifiers.

Header extension : Only one extension is allowed per packet. Its format is ID + length + data, with length measured in 32‑bit words. Extensions can be single‑byte or double‑byte.

Single‑byte header : Defined by profile 0xBEDE, IDs 1‑15, data length = len+1.

Double‑byte header : Defined by profile 0x100X, IDs 1‑255, data length = len.

Video Orientation (CVO) : Encoded in an 8‑bit field; the high‑order 4 bits are reserved, the next bit selects the camera mode, and the low‑order 3 bits represent one of eight possible rotation angles.

ffmpeg conversion : The extracted rotation angle can be applied with an ffmpeg filter, e.g., ffmpeg -i input.mp4 -vf "transpose=2" output.mp4 .

Summary : Parse CVO data from SDP, extract rotation from RTP header, map it to ffmpeg parameters, and perform the necessary video rotation during transcoding.