Spatio-Temporal Detection and Isolation: Results on PETS 2005 Datasets
Richard Souvenir, John N. Wright, Robert B. Pless · 2005
Recently developed statistical methods for background subtraction have made increasingly complicated environments amenable to automated analysis. Here we illustrate results for spatio-temporal background modeling, anomaly detection, shape description, and object localization on relevant parts of the PETS2005 data set. The results are analyzed both to distinguish between difficulties caused by different challenges within the data set, especially dropped frames, recovery time from camera motions, and what can be extracted with very weak assumptions about object appearance.