Accurate and automatic NOAA-AVHRR image navigation using a global contour matching approach

Abstract

The problem of precise and automatic AVHRR image navigation is tractable in theory, but has proved to be somewhat difficult in practice. Our work has been motivated by the need for a fully automatic and operational navigation system capable of geo-referencing NOAA-AVHRR images with high accuracy and without operator supervision. The proposed method is based on the simultaneous use of an orbital model and a contour matching approach. This last process, relying on an affine transformation model, is used to correct the errors caused by inaccuracies in orbit modeling, nonzero value for the spacecraft's roll, pitch and yaw, errors due to inaccuracies in the satellite positioning and failures in the satellite internal clock. The automatic global contour matching process is summarized as follows: i) Estimation of the gradient energy map (edges) in the sensed image and detection of the cloudless (reliable) areas in this map. ii) Initialization of the affine model parameters by minimizing the Euclidean distance between the reference and sensed images objects. iii) Simultaneous optimization of all reference image contours on the sensed image by energy minimization in the domain of the global transformation parameters. The process is iterated in a hierarchical way, reducing the parameter searching space at each iteration. The proposed image navigation algorithm has proved to be capable of geo-referencing a satellite image within 1 pixel.