Human behaviour recognition with segmented inertial data

摘要: The development and recent advancements of integrated inertial sensors has afforded substantive new possibilities for the acquisition study complex human motor skills ultimately their imitation within robotic systems. This paper describes continuing work on kinetic models that are derived through unsupervised learning from a continuous stream signals, including Euler angles accelerations in three spatial dimensions, acquired motions arm. An intrinsic classification algorithm, MML (Minimum Message Length encoding) is used to segment data, formulating GaussianMixture Model dynamic modes it represents. Subsequent representation analysis as FSM (Finite State Machines) found distinguishing consistent sequences persist across both, variety tasks well multiple candidates. exemplary “standard” sequence each behaviour can be abstracted corpus suitable data turn utilised together with alignment techniques identify behaviours sequences, detail homologous extent between each. progress contrast previous future objectives discussed.