Statistical Methods and Models for Video-Based Tracking, Modeling, and Recognition

摘要

Computer vision systems attempt to understand a scene and its components from mostly visual information. The geometry exhibited by the real world, the influence of material properties on scattering of incident light, and the process of imaging introduce constraints and properties that are key to interpreting scenes and recognizing objects, their structure and kinematics. In the presence of noisy observations and other uncertainties, computer vision algorithms make use of statistical methods for robust inference. In this monograph, we highlight the role of geometric constraints in statistical estimation methods, and how the interplay between geometry and statistics leads to the choice and design of algorithms for video-based tracking, modeling and recognition of objects. In particular, we illustrate the role of imaging, illumination,and motion constraints in classical vision problems such as tracking, structure from motion, metrology, activity analysis and recognition, and present appropriate statistical methods used in each of these problems.