Resumen de A unified framework for consistent 2d/3d foreground object detection
This Ph.D. dissemination addresses two-dimensional (2D) and three-dimensional (3D) active entity detection in video scenes. Active entities are the foreground parts in a stationary background scene and they typically correspond to the regions of interest in many applications such as automated video surveillance, object and person tracking and suspicious object detection, among others.
This dissemination presents a novel framework that permits obtaining 2D and 3D active entities as an inter-dependent probabilistic procedure. In the process of creating this framework, a study has been conducted to explore ways to generalize existing activity detection techniques to a Bayesian form. Some of the techniques, specially those which were closer to planar foreground detection, can be usually extended. With regard to volumetric activity detection, the literature reveals that very little work has been done in the field of Bayesian classification. Thus, in order to support the framework previously outlined, a new Bayesian 3D activity detection technique has been developed.A probabilistic analysis only accounts for half of the problem. The Bayesian framework gives a unified manner to interact between the planar and the volumetric detection tasks and helps to prevent the propagation of noisy pixel observations to the 3D space. However, when large systematic errors occur in the 2D detection level, a different approach has to be taken to correct them. In this respect, 2D/3D geometric relations can be exploited to detect systematic errors. Errors in the planar detection task often produce a set of incompatible foreground planar regions in the sense that they cannot be globally explained as the projection of the detected 3D volume. This is a key issue with significant implications that is not considered in most of current approaches.
