Walking Behavior Classification using Discrete Hidden Markov Models (Bachelorarbeit)
This thesis presents a probabilistic classification method for the recognition of walking behaviors. The widely used method of Hidden Markov Models is combined with a Code Book Quantizer to adapt to the new context of continuous pedestrian motion classification. The implemented system improves the functionality of indoor navigation systems and correct position error estimates, by providing additional context information. Using the foot-mounted unit of the PiNav-System, designed at the Institute for Real-Time Computer Systems (RCS), measurement data of walking behaviors are gathered from inertial measurement sensors, basically an accelerometer and a gyroscope. Processing raw data to assess characteristic information of walking behaviors, extracting significant streams as well as modeling, training and recognizing proceeded motions, are the main key issues in such systems and will be discussed in depth.