抄録
In this paper, we propose a robust online action recognition algorithm with a segmentation scheme that detects start and end points of action occurrences. Specifically, the alogorithm estimates reliably what kind of actions occurring at present time. The algorithm has following characteristics. (1) The algorithm incorporates human knowledge about relations between action names in order to toughen the recognition, thus it labels robustly multiple action names at the same time. (2) The algorithm uses time-series Action Probability that represents the likelihood of each action occurrence at every frame time. The Action Probability is obtained from time-series human motion using support vector machine. (3) The algorithm can detect robustly and immediately the segmental points using classification technique with hidden Markov models (HMIs) . The experimental results using real motion capture data show that our algorithm not only prevents the system from making unnecessary segments due to the error of time-series Action Probability but also decreases effectively the latency for detecting the segmental points.
