Metadata-Based Activity Analysis in 3D Tele-Immersion

In view of the necessity of activity analysis on Gbps-scale 3D Tele-Immersive (3DTI) content bundle and petabyte-scale 3DTI recordings, this work proposes and verifies the feasibility of replacing high-latency intrusive analysis with light-weighted metadata-based analysis. For real-time in-session use case, result shows that metadata-based analysis module, when personalized with user's body index, can achieve accuracies in 90 percentile on classifying various activity classes. For offline cross-session use case, we propose a hybrid analysis scheme which combines the advantages of intrusive analysis (i.e., conventional activity analysis on content level) and metadata-based analysis and achieves high accuracy with low computation latency.