In recent years, the surge in demand for remote work has increased the bandwidth requirements of video conferencing systems. While conventional video encoding standards perform well in high bandwidth environments, their quality and performance degrade significantly in low bandwidth and low bit rate conditions. To address this issue, semantic communication technology can be introduced, which reduces bandwidth demand by extracting and transmitting the semantic information of the video. Although extensive theoretical research has been conducted on semantic feature extraction and reconstruction for video, existing methods struggle with poor reconstruction quality during large head movements in practical applications. In this paper, we propose a semantic communication-based video conferencing system that improves the user experience in practical applications by introducing a preprocessing mechanism for video frames with large head pose. Additionally, we design and implement a testing platform to validate the system’s performance. First, we develop the system model and design a preprocessing module for video frames with large head pose. Then, we build a testing platform and conduct performance evaluations on the proposed system. Experimental results demonstrate that the proposed system successfully resolves the poor reconstruction issue during head movements in semantic video conferencing, while reducing the average bandwidth requirements by three-quarters compared to traditional video conferencing solutions.