Thesis (M.Sc.) - Cairo University - Faculty of Computers and Information - Department of Computer Science

Computer systems are major sources of problem solving in our daily life. The revolution of sensory technology and human-machine interfaces greatly affected research in how machine could detect and act properly against human actions. One of the most effec-tive approaches for human behavior analysis, simulation and understanding is modeling techniques. Researchers have introduced many approaches for modeling typical classes of behavior. Human behavior prediction and classi{uFB01}cation is also a core problem to be solved in any human behavior analysis model. Classi{uFB01}cation and prediction still suffer from a number of bottlenecks for selecting the best classi{uFB01}cation model that gives best classi{uFB01}cation results for a given data source. Fine-tuning classi{uFB01}cation parameters for a classi{uFB01}er to make best classi{uFB01}cation results in terms of generalization and classi{uFB01}cation accuracy is a known problem in research and is dependent on the data source being ex-perimented. In this work a set of modeling concepts were applied to design a behavior analysis model in terms of a set of data models and process models that incorporate together to build a system that should be able to qualify, quantify and analyze human behaviors. We also used biosignals as sources for human behavior learning and analysis in terms of electromyography (EMG) signals sensored over human skin and VICON data acquired from motion capturing systems. This research introduced a classi{uFB01}cation approaches that employed swarm intelligence for selecting best classi{uFB01}cation models and selecting most releveant features from a set of data sources

Issued also as CD