Title: Segmentation of Japanese rescue dog behavior data: A Markov-switching model approach
Authors: Rex Cheung - San Francisco State University (United States) [presenting]
Ryunosuke Hamada - Denso Corporation (Japan)
Abstract: The use of Markov Switching Vector Autoregressive Model (MSVAR) is proposed to analyze Japanese search and rescue dog behavior. As search and rescue dogs have been a popular companion in many police operations in many countries, they are rarely included in any search operations in Japan, due to the lack of data evidence to show the efficiency of the rescue dogs. Therefore, experiments have been done by mounting sensors and video camera to capture the movement signals of the dog, and to analyze the activity and behavior during a search operation. The aim is to apply the MSVAR model to analyze the sensor signals, by dividing the observed signal into homogenous regions such that each region corresponds to a unique, non-overlapping activity of the dog (such as run, sniff, etc.). To encourage flexibility, a sparse-lag penalty will be imposed to perform automatic lag selection and sparsification of autoregressive coefficients.