| Abstract |
The analysis of image sequences containing moving non-rigid objects, such as people, has been the subject of considerable research in the field of computer vision, pattern recognition, and neural networks. The human tracking is a difficult problem because of unpredictable and complicated human motion and it is one of the most interesting problems in this field of science. Tracking points in the human body could help 3D reconstruction and recognition of human movement. First, we tried to solve the problem of human parts and 18 basic points extraction using POSTURE. Our purpose is to develop a fast and efficient algorithm. In this problem, there are many solutions because of projection from 3D in 2D image plane. Our method provides only one of them. Furthermore, we don't use color information. The first stage of the method is the 2D human part's segmentation and the second stage is the 18 basic human points localization. Also, we developed an original method to solve the 2D human tracking problem using the 18 major joint human points. The human body is divided geometrically into 6 main parts and we track each part separately. We used a 2D model, a prediction method, geometry constraints of human body and color information. The algorithm demands the position of human in the first image of sequence. Our basic purpose is to develop a low computation cost algorithm that can be used independently of camera motion. Also our method recognizes if any human part is visible or not. The outputs of the algorithm are the position of 18 basic human points and a 2D human segmentation in every image of video sequence. Both methods' precision is satisfactory. However, there are cases where the tracking algorithm fails because of low quality sequences or low background contrast. These cases can be recognized by high matching errors.
|
Αναγνώριση και Παρακολούθηση των Μελών του Ανθρωπίνου Σώματος σε Κίνηση
