Human behaviour recognition in data-scarce domains
作者：
Highlights：
• We challenge the notion that the exact temporal structure of activities needs to be modelled.
• We compare performance against a Hidden Markov Model baseline.
• The weak temporal structure of our approach makes it less sensitive to observation order.
• Hidden Markov Models cannot be used to classify some activity sequences.
• Our approach outperforms the baseline by 17%.
摘要
Highlights•We challenge the notion that the exact temporal structure of activities needs to be modelled.•We compare performance against a Hidden Markov Model baseline.•The weak temporal structure of our approach makes it less sensitive to observation order.•Hidden Markov Models cannot be used to classify some activity sequences.•Our approach outperforms the baseline by 17%.
论文关键词：Behavior recognition,Bayesian inference,Visual surveillance,Behavior decomposition
论文评审过程：Received 20 December 2013, Revised 25 November 2014, Accepted 19 February 2015, Available online 6 March 2015.
论文官网地址：https://doi.org/10.1016/j.patcog.2015.02.019
