Adaptive image processing and sensing technologies for detecting anomalous conditions near critical transportation infrastructure assets

This work presents a methodology for detecting anomalous conditions on transportation infrastructure assets and generating alarms to draw the attention of personnel monitoring them via various sensing technologies. The methodology developed, which consists of a four part process, is capable of detecting conditions that vary substantially from those that are normally observed. This four part process consists of the following steps: (1) an intensity characteristic model of the transportation asset is learned during a training phase. (2) During the foreground object segmentation phase, objects are segmented based on pixel characteristics that vary substantially from those embodied in the intensity characteristic model. (3) Segmented objects that match certain morphological, topological, and/or geometric constraints are flagged as being candidates for further (temporal) processing. (4) Segmented objects with unanticipated temporal persistence or geometric characteristics are then identified and brought to the attention of a human operator.