P. Spagnolo, A. Branc, G. Attolico, and A. Distante (Italy)
Visual Surveillance, Motion Detection, Background Subtraction, Shadow Elimination
Visual surveillance of outdoor environments is one of
the most fundamental applications of computer vision.
Our context is the visual surveillance of archaeological
sites: the goal is the extraction of moving objects, their
classification and the interpretation of gestures of the
detected human figures, in order to discriminate illegal
actions from legal ones.
In particular, we focus on the core problem of detecting
and extracting moving objects and of eliminating the
shadows. The latter step is crucial because shadows can
create difficulties to any classifier charged of correctly
discriminating the observed moving objects as people,
cars, animals, etc.
In this work we propose an improvement of a well
known algorithm for the detection of moving object,
that makes it more robust with respect to different noisy
Moreover, we propose a new approach for removing
shadows cast by moving objects. Our idea is to detect
shadow regions as background regions that have not
substantially changed their texture. We discriminate
shadows from moving objects by comparing the
photometric gains of neighbouring pixels.
The system has been tested on real sequences acquired
using a fixed camera into an archaeological site.