Abstract
Digital Surface Models (DSMs) generated from satellite stereo imagery provide valuable but not comprehensive information for building change detection. Therefore, belief functions have been introduced to solve this problem by fusing DSM information with changes extracted from images. However, miss-detection can not be avoided if the DSMs are containing large region of wrong height values. A refined workflow is thereby proposed by adopting the initial disparity map to generate a reliability map. This reliability map is then built in the fusion model. The reliability map has been tested in both Dempster-Shafer Theory (DST), and Dezert-Smarandache Theory (DSmT) frameworks. The results have been validated by comparing to the manually extracted change reference mask.
Highlights
In our previous research [1] [2], belief functions have performed very well for 3D building change detection
To circumvent the problems of DS rule, Smarandache and Dezert, Martin and Osswald have developed in Dezert-Smarandache Theory (DSmT) [5] two fusion rules called PCR5 and Proportional Conflict Redistribution Rule #6 (PCR6) based on the proportional conflict redistribution (PCR) principle which consists
Our previous research has evidenced the performance of the belief functions in Digital Surface Models (DSMs) assisted change detection [2]
Summary
In our previous research [1] [2], belief functions have performed very well for 3D building change detection. The DSMs may still exhibit some outliers resulting in occlusions within the stereo/multi views and due to matching mistakes. In this case, change information from spectral information of the original stereo imagery can and should be used together with height changes to eventually highlight building changes. Change information from spectral information of the original stereo imagery can and should be used together with height changes to eventually highlight building changes For this purpose proper fusion theories and approaches are needed. The belief functions and building change detection fusion models are
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have