STATISTICS-BASED FUSION OF TERRAIN DATA SETS AND CHANGE DETECTION Jen-Jer JAW

This study aims at outlining a working experience where fusion and change detection of terrain data sets of dual-temporal but of the same area was performed via the technique of least squares adjustment. The proposed working flow first predicted the heights of one of the terrain data set onto the other one by means of least squares collocation, leading to point to point correspondence of the data sets on a horizontal component basis. The varied heights on the same horizontal positions in two data sets were considered the data type of double measurements of the very same terrain but with different variance-covariance matrices. A thresholding process followed by considering the randomness of the data sets, resulting in highlighting those points with change potential. The adjustment using weighting average then fused the heights that were reported unchanged. The areas of change were detected by the method of clustering. The outcomes in this study showed not only the higher accuracy of the fused terrain but also the areas of change in concern. Last but not least, the proposed workflow did not require any specific data format, thus avoiding degrading the original data quality due to unnecessary interpolation processes like other methods do.