Track anomaly detection with rhythm of life and bulk activity modeling

This paper describes a model and algorithm for detecting anomalies in track data. The algorithm is general in the sense that it can be applied to tracks from any type of sensor. Two important enhancements to the standard algorithm are outlined. The first of these characterizes predictable temporal variations in behavior, the so-called rhythm of life. This allows the detection of unusual activity during one part of the day that would be considered normal at other times. The second development analyzes the bulk behavior of targets. Although tracks on their own may not be unusual, their combined actions could be suspicious. Also, bulk activity analysis allows the detection of missing expected behavior, which is not possible with many techniques. Application of the algorithms is demonstrated by analyzing the movement of people in a canteen.