Abstract
Abstract. Digital elevation models are one of the basic products that can be generated from remotely sensed imagery. The Semi Global Matching (SGM) algorithm is a robust and practical algorithm for dense image matching. The connection between SGM and Belief Propagation was recently developed, and based on that improvements such as correction of over-counting the data term, and a new confidence measure have been proposed. Later the MGM algorithm has been proposed, it aims at improving the regularization step of SGM, but has only been evaluated on the Middlebury stereo benchmark so far. This paper evaluates these proposed improvements on the ISPRS satellite stereo benchmark, using a Pleiades Triplet and a Cartosat-1 Stereo pair. The over-counting correction slightly improves matching density, at the expense of adding a few outliers. The MGM cost aggregation shows leads to a slight increase of accuracy.
Highlights
Creation of digital elevation models by automatic image matching of airborne or spaceborne optical data is one of the basic procedures in photogrammetry
The test region in Catalonia, near Barcelona has been selected due to the availability of several stereo satellite datasets and a good reference data set provided by the Institut Cartografic de Catalunya (ICC)
The datasets were bundle adjusted using tie points and ground control provided by the ICC
Summary
Creation of digital elevation models by automatic image matching of airborne or spaceborne optical data is one of the basic procedures in photogrammetry. The Semi-Global Matching (SGM) algorithm (Hirschmuller, 2008) has been successfully applied to a variety of stereo problems. 2. RECENT SGM IMPROVEMENTS pixels in the image pair. In a thorough evaluation of many matching cost functions (Hirschmuller and Scharstein, 2009), Census turned out to be a very robust and reliable cost function with good performance. As the matching costs based on single pixel values or small windows are ambiguous, regularization is used to ensure a well behaved reconstruction. (Facciolo et al, 2015) proposes improvements to the aggregation algorithm Their contributions were evaluated on Middlebury close range data (Scharstein et al, 2014), with often has different properties than close range data
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
