Figures & data
Figure 1. Aerial map from Bing Maps© with blue rectangles highlighting zones that have been recently transformed.
![Figure 1. Aerial map from Bing Maps© with blue rectangles highlighting zones that have been recently transformed.](/cms/asset/c30a5888-41e6-4b74-b25d-ab9b86d808a1/tgsi_a_1244998_f0001_oc.gif)
Figure 5. Example of top-down view localization in the aerial image. The top-down view T (a) is compared with the large aerial image A (b), and the green area denotes the found localization of T.
![Figure 5. Example of top-down view localization in the aerial image. The top-down view T (a) is compared with the large aerial image A (b), and the green area denotes the found localization of T.](/cms/asset/b58ab1a4-7151-4bd5-ade9-ee1585fe97ac/tgsi_a_1244998_f0005_oc.gif)
Figure 8. Artifacts brought by built-up structures (from top to bottom): missed changed, false changes due to other structured objects, false changes due to deformation brought by top-down view. For each line are given (from left to right): ground panorama, top-down view, and corresponding aerial image.
![Figure 8. Artifacts brought by built-up structures (from top to bottom): missed changed, false changes due to other structured objects, false changes due to deformation brought by top-down view. For each line are given (from left to right): ground panorama, top-down view, and corresponding aerial image.](/cms/asset/ab0cd1d2-f184-41e6-a53c-81051f7c1ffd/tgsi_a_1244998_f0008_oc.gif)
Table 1. Confusion matrix for the changed/unchanged classification.
Table 2. Confusion matrix for the unchanged/changed/structured classification.
Table 3. Computational complexity of the different steps composing our proposed approach. CPU times have been averaged among 100 runs.