Temporal structure methods for image-based change analysis

This paper addresses the exploitation of massive numbers of image-derived change detections. We use the term ldquochange analysisrdquo to emphasize the intelligence value obtained from large numbers of change detection over long time intervals, rather than the emphasis by most researchers to date on ldquochange detectionrdquo methods and small numbers of change detections. Our methods emphasize local temporal descriptions of activities and include minimal spatial information about activities. Our three methods adapt and extend: (1) classic unsupervised pattern recognition operating on bag-of-words features; (2) Latent Semantic Analysis (LSA); and (3) probabilistic LSA (PLSA). These methods allow us to: (a) Detect and describe anomalous activities; (b) Discover categories of activity, describe a category of activity, and assign an activity to a category; (c) Retrieve similar activities from a historical database. We present experimental results that compare our methods (1)-(3) for performing functions (a)-(c), using webcam images of a town market square collected every few minutes over 74 days. We discuss how our techniques are equally applicable for change analysis using wide-area sensors.