Hidden Compensation and Shadow Enhancement for True Orthophoto Generation
4. Case Study
To realize the algorithms presented in previous sections, a fully automatic orthographic rectification procedure was developed. Along with the source images, inputs of the procedure include DBM, existing DTM, pre-calculated orientation parameters and direction of the sun. Among the cases studied, the quadruplet covering Fu-Zen University is chosen to illustrate the important features of the proposed scheme. Figure 4 shows the photo data from which the rectangular region highlighted in white box was picked for inspection in the following discussion. The scale of the quadruplet is around , the overlap along and across trip are both about . The photo data was scanned at a spacing of to produce digital images with nominal ground resolution. By assuming the shape of most buildings may be decomposed into vertical walls and horizontal or slant rooftops, the DBM was constructed numerically from coordinates of all building corners measured from analytical plotter.
Figure 4, The quadruplet under studying.
For comparison, fig.5 shows for the highlighted area the conventional orthophoto generated from the lower-left and the upper-right scenes. It demonstrates that without hidden detection and relief displacement correction for buildings, locating from these products the right position of a given object would be difficult. It is also problematic for visual interpreter to decide whether a particular feature shown on these products should be appeared. Impacts of shadow defect are also distinctly observed in this figure. Besides the difficulty of recognizing objects in shadow areas, it is sometime a challenge task just distinguishing shadow itself from the other ground features.
Figure 5,Conventional orthophotos from lower-left and upper-right photo.
With the upper-right scene as the master image, a true orthophoto was generated using the procedure developed in this study. The intermediate results and the final product for the highlighted area are shown in fig.6. Fig.6a shows result with hidden detection and relief displacement correction applied. The hidden areas were left blank to avoid possible confusion. Locating from this product the accurate position for viewable object is possible. As shown in fig.6b, the hidden defect has been significantly reduced after a fill-in operation. Still few small hidden areas exist. For all the master and slave images, the highlighted area is located right to the central line. Chance is that some part of ground features west to tall buildings is hidden from either of the images. A close look at fig.6b, unnatural discontinuity may be observed along the fill-in boundary. This is improved as shown in fig.6c by substituting direct fill-in operation with the seamless mosaic technique. Applying further the local enhancement for shadow defect, the final true orthophoto product as shown in fig.6d is with improvements, especially in the accessibility for objects obscured in shadow areas. Compared to fig.5, the interpretability of fig.6d is significantly improved.
Figure 6, Intermediate and final products.
(a) Relief displacement correction only.
(b) Hidden compensation applied without smoothing.
(c) Hidden compensation using seamless mosaic technique.
(d) Shadow enhancement applied.
