Real-time video communication is very important to the Internet applications of video conferencing, video telephony, video-on-demand, and etc. However, the heavy traffic of video data along with its timing constraints makes it a challenge to provide large-scale, high QoS media streaming service over the current best-effect network. This paper presents a self-adapted feedback algorithm for video transmission which is based on the FEC coding technology and the KALMAN filter theory. By modifying the rate using the adapted KALMAN filter, this algorithm efficiently solves the problem of rate fluctuation caused by the lost packets. It reduces the network load caused by the retransmission of lost packets. It also weakens the influence of the loss of elementary layer packets (or other important contents) in transmission and provides the continuity of the video transmission over Internet. Simulation results indicate that this algorithm can guarantee satisfied performance for video transmission in the high-packet- lost networks.