Digital Object Identifier (DOI) 10.1007/s00138-003-0122-5 Machine Vision and Applications (2003) 14: 197–198 Machine Vision and Applications Editorial Introduction to the special issue on human modeling, analysis, and synthesis Published online: 8 August 2003 – c  Springer-Verlag 2003 Welcome to this special issue on human modeling, analysis, and synthesis of the international journal Machine Vision and Applications! The last decade has witnessed an explosion of research on new ways of understanding and representing the subtleties of human modeling, analysis, and synthesis. A se- ries of successful workshops and conference tracks in this area further exemplify the vital role of capturing information about humans using video cameras. It is an exciting and challenging research topic, and researchers in computer vision and many other interdisciplinary areas are making good progress in solv- ing a number of long-standing issues. In pursuit of a better un- derstanding of the problems associated with this theme, this special issue seeks to put together selected papers on aspects of modeling, analysis, and synthesis of a person’s appearance, movement, and behavior. The papers appearing in this spe- cial issue were carefully reviewed by a panel of experts in the relevant areas. The domain of analysis and synthesis of human motion re- quires complex image-understanding tools involving a number of areas (e.g., geometry, dynamics, complex time evolution) and has numerous applications. For example, motion measure- ments in medicine can be used to characterize and diagnose disorders caused by disease or injury and to assess the effi- cacy of drugs/therapies used to treat the problem. Also, 3D motion measurement, biomechanics, and real-time biofeed- back are modern techniques used by sports scientists to rapidly improve the performance of professional and Olympic ath- letes. Motion measurement has many other applications such as worker training, worker physical baseline testing, robotics, product design, telepresence, and even crash testing. How- ever, the processing of human motion presents challenges due to strong viewpoint variabilities of the visual motion patterns. By incorporating powerful domain knowledge, model-based approaches are able to overcome these problems to a great extent and extract meaningful information for many applica- tions. For example, surveillance applications may require the detection of humans or the analysis of meaningful motion for recognition of unusual behaviors (e.g., intention behind purpo- sive and communicative behavior). Similarly, recognition for biometric authentication requires understanding of the tem- poral aspects of human motion and the meaningful fusion of human body motion information with semantic knowledge. Hence, image-understanding tools, which can cope with these difficulties to meet the needs of real applications, are of great importance. The first part of this special issue includes position state- ments from five researchers (i.e., J. Aggarwal, B. Dariush, A. Hilton, M. Shah, and M. Trivedi) who participated in the IEEE Workshop on Human Modeling, Analysis and Synthe- sis on Hilton Head Island, South Carolina, June 16, 2000. We thank them for their important contribution. The second part includes the papers selected. Specifically, the first paper, “Automatic acquisition and initialization of articulated models”, by N. Krahnstoever et al., addresses the problems related to automatic acquisition and initialization of articulated models from monocular video without any a priori knowledge. The kinematic structure and shape of complex articulated objects was extracted from video sequences and is subsequently used to obtain the information to build corresponding articulated models. It was argued that 3D models can be acquired in a similar fashion if two or more simultaneous camera views are available. The important problem of estimating an individual’s an- thropometry and pose from a single uncalibrated image is ad- dressed in the second paper, “On the improvement of anthro- pometry and pose estimation from a single uncalibrated im- age”, by C. Barron and I.A. Kakadiaris. The proposed method effectively exploits prior statistical anthropometric informa- tion to constrain the estimation process and allows the simul- taneous estimation of both anthropometry and pose from a single image. The third paper in this context is “Modeling and estimating the pose of a human arm”, by T.B. Moeslund and E. Granum. The key idea here is to match the pose of a human arm with a silhouette image by noting constraints, which dramatically reduce the fitting search space. The paper addresses the issue of pose estimation of the upper and lower arm of a human from a monocular image sequence. The kinematic constraints eliminate all unattainable configurations of the human arm and thereby reduce the total number of different “possible” poses significantly. The pose is estimated using an analysis by synthesis approach, that matches the model silhouette with the real silhouette.