Real-time Pattern Isolation and Recognition Over Immersive Sensor Data Streams

Data streams appear in many recent applications, where data are constantly changing or take the form of continuously arriving streams. We focus on data streams generated by sensors for monitoring users in immersive environments. To recognize users' interactions, we need to analyze the aggregation of several sensor data streams and match the result to a set of known actions. In addition, we need to separate a continuous series of actions into recognizable atomic actions. Hence, we first propose a distance metric, weighted-sum Singular Value Decomposition (SVD), suitable for similarity measurement of immersive data sequences. Subsequently, we propose a mutual information based heuristic for separation of the action sequences. Finally, we perform several empirical experiments using real-world virtual-reality devices to verify the effectiveness of our approach.