In view of the high resource demand of 3D Tele-immersion (3DTI), we propose a user activity-based resource adaptation scheme that adjusts the compression ratio of 3DTI videos according to the user activity. The compression technique we designed is based on the frame synthesis via feature-based morphing. This technique is customized for 3DTI video due to the special properties of its scenes and the depth information provided by the system. Via a machine learning approach, our system can classify the user's current activity and choose the most suitable priority among temporal resolution, spatial resolution, and resource consumption in the adaptation. Our results show that the resource saving can reach up to 25% without perceptible degradation of the 3DTI videos.