Summary: | 碩士 === 國立中央大學 === 土木工程研究所 === 94 === In order to make more realist visualization of 3D building models, the generated scenes should have texture of building facades in addition to the geometry of buildings. Using video sequences as the sources of realist texture is an effective approach. However, a few issues must be addressed in using video sequences for texture mapping in 3D visualization. If a single video frame can not cover the entire wall, individual frames need to be merged correctly to generate a complete image. When foreign objects, like trees, block the target facade, they need to be removed and the blocked areas should be recovered. Also, complete video mosaics need to be mapped onto corresponding building models.
This study developed a procedure to create texture mosaicking images semi-automatically from close-ranged video sequences for photo-realistic texture mapping of building models in 3D visualization applications. The first step was to correct geometry of images. It not only improved the mosaicking, but also facilitated the mapping transformation. The mosaicking was done using interest points detected with corner detecting algorithms and matched with distance-filtering and Normalized Cross Correlation operator. In the developed system, image morphology was used to identity mirroring axis for recovering blocked areas automatically. The resultant building models should have more accurate texture information and improved the reality and practicality of cyber city implementation
|