Change detection in temporally related image sequences is a primary tool for extraction and detection of activities in background scene with vast and wide range of applications ranging from security and surveillance to fault detection and power savings. The prevalent methods for change detection are derived from the difference extraction where differences in the gray-level of values of the pixels between the two or more image sequences are used for the estimation and prediction of these changes. However this approach and its derived modifications are largely dependent and reliant on the application of value thresholds to provide significance to the differences, in order to compensate for the vulnerability of these methods to illumination variability and noise. A frequency domain approach to change detection is proposed that eliminates the need for thresholds and provides comparatively superior performance to the existing algorithms.
展开▼
