|
Abstract:
|
A method is proposed for the detection and estimation of change points in an anisotropic Markov random field of events occurring on a surface. The naturally appearing ten canonical parameters reflect the overall intensity and the dynamics of events in terms of the spatial correlation, speed, and the direction of spread. Each parameter and its estimator has a specific meaning that describes the development of events and connections between neighboring events occurring close to each other on a spatio-temporal scale. A threshold is chosen to control the rate of false alarms triggered by the change-point detector. When a significant change is detected, it may indicate a breakdown, a potential problem, or even a threat such as organized crime. Applications are shown in semiconductor industry and in threat detection.
|
