Reference-less Human Motion Recognition using MEMS- based inertial motion sensors and stochastic signal modelling

摘要

Human Motion is a highly variable and multidimensional form of displacement and rotation series in space performed by multiple parts of a moving body, i.e. different muscles, bones and joints working together. As adult humans have mastered to optimize different movements in early childhood (learning from other people and/or from own mistakes), these movements seem obvious to them in everyday life and hence evoke no need for further query or perfection. In professional sports or in applications of rehabilitation and advanced training a reliable possibility of computer-assisted motion analysis and validation can be a key for optimized training procedures and success measurement. The present work shows the latest research results performed at the CCASS aiming for providing a framework for reference-less human motion analysis and validation using low-cost inertial motion sensors and a light-weight, full-body mutli-sensor suit. The developed algorithms base on the theory of Hidden Markov Models and on stochastical modelling of human motion using Markov chains. In the present paper the motion recognition concept will be explained as well as the model definition, the feature selection and the validation results will be discussed. Ultimately, impressions from the sensor suit development and the future work will be given.