Incremental estimation of image-flow using a Kalman filter

Abstract

Many applications of visual motion, such as navigation, tracking, etc., require that image-flow be estimated in an on-line, incremental fashion. Kalman filtering provides a robust and efficient mechanism to record image-flow estimates along with their uncertainty and to integrate new measurements with the existing estimates. The fundamental form of motion information in time-varying imagery (conservation information) is recovered along with its uncertainty from a pair of images using a correlation-based approach. As more images are acquired, this information is integrated temporally and spatially using a Kalman filter. The uncertainty in the estimates decreases with the progress of time. This framework is shown to behave very well at the discontinuities of the flow-field. Algorithms based on this framework are used to recover image-flow from a variety of image-sequences.<>