Monitoring Slow-Moving Landslides Using Spatially Adaptive Least Squares Image Matching

Abstract

This study presents and evaluates an algorithm which uses the least squares image matching (LSM) with spatially adaptive and high-order geometric models to estimate horizontal surface displacements of slow-moving landslides from repeat optical images. Pairs of high resolution optical images over a rockglacier creep and a slow-moving landslide are orthorectified and co-registered. Image matching is applied first using the conventional normalised cross-correlation (NCC), then using the LSM algorithm with image-wide single geometric models, and finally using the LSM with spatially adaptive geometric models. The spatially adaptive algorithm operates in such a way that for each template the model that produces the lowest sum of square of intensity difference is considered the best fitting model. Affine, projective and second-degree polynomial geometric models are included. The algorithms are evaluated in reference to the NCC algorithm based on the signal-to-noise ratio of reconstructing the reference image from the search image.