| As an important way of communication in people’s work and life,video conferencing systems are widely used in educational institutions,business units,government agencies and other occasions.Traditional online video conferences integrate the video streams of all participants in a flat window.This method lacks a multi-dimensional experience similar to real offline meetings,and the interaction and collaboration between participants is greatly reduced.Note It is easy to lose strength and lack a certain sense of immersion and participation.Therefore,the research on the video scene fusion of the virtual meeting room system under the three-dimensional scene and the provision of three-dimensional solutions for the video conference will have very important research significance.Aiming at the problems of how the 3D virtual meeting room system processes the video stream in real-time and fusion with the 3D scene under the condition of Non-GPU,this paper proposes a two-stage processing fusion scheme based on kernel correlation filter tracking,Gaussian mixture model segmentation and video stream texture mapping..First,the original video stream is preprocessed,and the first frame of the video stream is processed through a lightweight target detection network as the input for subsequent tracking;the core-related filtering algorithm is used to track the preprocessed video stream,and the cyclic matrix is used to simplify the calculation to improve tracking Speed:Establish a Gaussian mixture model for the video sequence to describe the pixel distribution,use the interactive iterative process of continuous model parameter learning and segmentation estimation to minimize energy segmentation,improve segmentation accuracy,and scale the image sequence to ensure real-time performance.Then,the processed video stream undergoes texture transformation through canvas elements,and the transformed video stream texture object is mapped to a three-dimensional scene in a mapping manner to achieve a video scene fusion effect.In order to improve the Qo S of the system,starting from the server streaming media quality and network control,a multi-connection streaming media transcoding scheme and a congestion control algorithm based on delay and packet loss are proposed.Through the experimental evaluation of the algorithm and system performance in the public data set,the experimental results show that the long-term target tracking and segmentation plug-in tracking speed of the video stream can be maintained at 60 FPS,the segmentation accuracy rate is over 86%,and the segmentation time is only 36 ms,reaching the system’s Performance requirements.At the same time,the system CPU and memory usage rate of the multi-connection streaming media transcoding solution is far lower than that of the ordinary encoding solution.The network delay,network jitter,and packet loss rate of the 3D virtual conference room system also meet the parameter requirements of video conferences. |
