Human Modeling, Reconstruction and Recognition in Video

Final Activity Report Summary - HUMOREV (Human Modeling, Reconstruction and Recognition in Video)

Detecting humans, reconstructing their structure and motion or understanding human action from video sequences are problems of key importance in the advancement of a variety of technological fields including video indexing and coding, entertainment and culture, animation and virtual reality, intelligent human-computer interfaces, protection and security. HUMOREV has created technologies to enable the construction of such systems by leveraging large scale statistical modelling and learning methods, optimization techniques and computer vision algorithms.

For efficiency and robustness, we have developed methods for the automatic learning of human model representations based on latent variables that are adapted (focused or profiled) for the task the human performs, and designed robust visual reconstruction and recognition methods that have integrated bottom-up discriminative human segmentation, detection and action prediction methods with top-down, complex structural human models.

These results have been combined in two integrated systems: one that can reconstruct the structure, motion and appearance of interacting humans in static images and video, and the other that recognises human activities in real world image sequences filmed in offices or outdoor environments. Demonstrators and graphical illustrations showing our human sensing capabilities as well as a summary of the work performed is available online at: http://sminchisescu.ins.uni-bonn.de/rep/mcextp-final.html.

Final Activity Report Summary - HUMOREV (Human Modeling, Reconstruction and Recognition in Video)

Diese Seite teilen

Herunterladen