A real-time system for video surveillance of unattended outdoor environments

This paper describes a visual surveillance system for remote monitoring of unattended outdoor environments. The system, which works in real time, is able to detect, localize, track, and classify multiple objects moving in a surveilled area. The object classification task is based on a statistical morphological operator, the statistical pecstrum (called specstrum), which is invariant to translations, rotations, and scale variations, and it is robust to noise. Classification is performed by matching the specstrum extracted from each detected object with the specstra extracted from multiple views of different real object models contained in a large database. Outdoor images are used to test the system in real functioning conditions. Performances about good classification percentage, false and missed alarms, viewpoint invariance, noise robustness, and processing time are evaluated.