| The ability to create three-dimensional (3D) texel models, using registered texel images (fused ladar and digital imagery), is an important topic in remote sensing. Offering both visual and dimensional accuracy, texel models are becoming invaluable tools for disaster management, situational awareness, and even automatic target recognition (ATR). These models are automatically generated by matching multiple texel images into a single common reference frame. However, rendering a sequence of independently registered texel images often provides challenges, and the problems that arise when viewing texel models will be discussed and corrected in this paper. As a result of stretched triangles and sensor noise, standard rendering techniques are not suitable for texturing 3D texel models. Although accurately registered, the model textures are often incorrectly overlapped and interwoven. Forming a texture atlas (stitching images together) is impractical, due to the independent nature of the underlying data sets. Consequently, corrections must be done after all the primitives have been rendered by determining the best texture for any viewable fragment in the model. Determining the best texture is difficult, as each texel image remains independent after registration. The depth data is not merged to form a single 3D mesh, thus the individual image textures must be rendered separately. It is therefore necessary to determine which textures are overlapping and how to best combine them during the render process. The best texture for a particular pixel can be computed using 3D geometric criteria, in conjunction with a real-time, view-dependent ranking algorithm. As a result, overlapping texture fragments can be hidden, exposed, or blended according to their computed measure of reliability. The visual defects of a texel model are reduced when overlapping textures are rendered using the view-dependent algorithm discussed in this paper. Successful rendering examples using synthetic data sets are presented. |
