New approaches to solve encountered problems in tracking of human movements

The development of data processing capacity of computers, the production of advanced but inexpensive cameras and the requirement of use of visual tracking systems in many areas of daily life depending on the emergence of the needs such as visual tracking, monitoring and managing have increased interest of researchers in the object tracking issue in recent years.In this thesis, three novel visual tracking methods, which can provide solutions to three different real world problems, are presented. In the first tracking method, the tracking people wandering in crowded scenes is intended. The developed multiple human tracking method assigns a unique identity number to each person in view of fixed video camera angle. Being able to adapt to varying the light amount and dynamic background information of environment, the proposed multiple human tracking method could maintain the tracking process for a high success rate in unintended conditions such as exposure to partial occlusion of human view.The second tracking method is the statistical human action recognition method based on skin detection. This realized method provided support to treatment of Alzheimer's patients. By means of the proposed tracking method, it is determined that Alzheimer's patients, who must use drug at regular intervals, use the given drug in time or not. The information whether the patients received the drugs or not, is obtained after visual tracking hands and faces of them. The proposed tracking method gives the right decision at approximately 92% success ratio.In the third tracking method, 3-D body pose estimation of a moving person is performed by using 2-D image sequences obtained with the help of a single camera. Pose estimation method performed is based on estimates of body joint angles by taking the projection of 3-D model of an artificial person on 2-D image. To estimate current body joint angles, the color and shape information of each body segment are combined under a Bayesian framework. A simulation of 3-D artificial human body obtained by using body joint angles is performed. The movements of ?walking?, ?running? and ?jumping? are tracked successfully with the help of the 3-D artificial human model obtained.