Although computer vision has still rather low acceptance as a reliable tool
for automated analysis of general human behavior it can very useful for
automated monitoring of relatively structured processes on a 24/7 basis even
under adverse conditions. This presentation deals with the case study of
human behavior analysis on an automobile production line, where noise and
occlusions pose serious challenges to the tracking algorithms. We examine
issues such as feature extraction, modeling of noisy time series,
multi-camera fusion, extraction of events in real time and active learning
in time series using these challenging videos.
Part of this work has been conducted in the framework of the SCOVIS project
(www.scovis.eu).