Model-Based Human Motion Analysis

The domain of human motion analysis has been widely studied among many researchers,with each one targeting different aspects of it i.e.human behavior understanding,pose inference of a person,estimation of forces acting on the body,matching and so on.However,dynamic nature of human being and the surrounding environment causes great deal of issues,such as random movements,clothing based variation,and occlusions which makes it difficult for the analysis to be successful.In respect to the mentioned problems,this dissertation presents the following individual contributions to ease the process of human motion analysis:1)Presented a deformable ellipsoidal based human limb model that can illustrate the muscle behavior of the limb.The deformation is intuitive and easily controllable which makes this approach attractive for researchers seeking model matching and force interaction on human skin.2)Demonstrated a new method of creating complex human limb parts using tensor product surface,along with a unique curved based deformation technique which will help significantly in reconstruction of human body parts from provided data set.3)Provided a human pose-based body parsing methodology that decompose the detected human into main identified body parts.A combination of pose-specific graph cut estimation and negative curvature minima cuts to provide detail segmentations,which is essential requirement for understanding human movements and behavior.The discussed algorithms and methodologies are relevant for markerless-vision based systems.From the experimental results obtained and approaches discussed in this dissertation we concluded that this work introduces new ways of dealing with multiple human motion analysis issues and the accuracy along with the overall performance is significant to complement the analysis.