Texture generation and mapping using video sequences for 3D building models

Three-dimensional (3D) building model is one of the most important components in a cyber city implementation and application. This study developed an effective and highly automated system to generate and map (near) photo-realistic texture attributes onto 3D building models using digital video sequences. The system extracted frames with overlapped textures of building facades and integrated them to produce complete texture images. Interest points on the extracted video frames were identified using cornerdetectors and matched with normalized cross-correlation for seamless stitching. Shadows and foreign objects were identified and removed with morphological algorithms and mended by mirroring neighborhood textures. Completed mosaicked texture images were mapped onto corresponding model facets by linear or parametric transformation. Test examples demonstrate that the developed system can effective generate seamless photo-realistic texture images and correctly map them onto complicated 3D building models with high efficiency.