Research On Human Body Segmentation In Visual Human Activity Recognition

Visual Human Activity Recognition is a leading technology in the field of computer vision in recent years, and human body object segmentation is a critical problem in the research. The results of segmentation play important roles in object recognition and behavior understanding. The research on human segmentation can not only make the application of human activity recognition break through the using situations "bottle-neck", but also the research results can be used in Video Retrieval, Video Compression and other fields the technique is needed in.The visual human activity recognition involves capturing single or multiple video sequences, tracking human body, deriving human motion data, recovering human 3D motion or describing and understanding human activities. It has a wide spectrum of promising applications in visual surveillance, sport motion analysis, 3D cartoons, games, virtual reality, user-interface and clinical studies.The 3D recovery and behavior understanding in studio or simple situation have been realized. But in complex situations, detecting and tracking multiple people or their body parts are usually implemented by means of special imaging devices. This approach can tackle occlusion, determine whether a person is carrying an object. But in these researches, segmentation problem is commonly avoided, because of lacking the efficient human body segmentation methods in complex situations.The current moving human body segmentation algorithms are often designed for fix situations or given tasks, and can not work adaptively when human position, attitude and environment change.This thesis focuses on the problems of human segmentation in visual human activity recognition, attempts to research the approaches to segment human objects in complex situations. The complex situation includes any condition of weather or illumination, the disturbance of shadow and occlusion. The human body feature description, the adaptive feature selection and segmentation algorithms based on features are also studied.The aim of this thesis is to study the human body segmentation technique in complex situations, give the approaches to deal with complex background changing,...