Detection of unusual human activity based on sequence of actions with MHI and CDP
Hiroki Murayama, Keiichi Yamada · 2010
This paper presents a method for detecting unusual human activity from a video stream, depending on whether a sequence of human actions differs from the usually observed sequences of human actions. The proposed method extracts low-level features from a video stream during a short time period for describing human actions, and the extracted sequence of the low-level features is used for describing human activity. Histogram of the oriented gradients of Motion History Image (MHI) is used for the low-level features, and Continuous Dynamic Programming (CDP) is used for calculating the similarity between the sequences. The effectiveness of the proposed method is demonstrated by an experiment conducted in an office.