摘要
为从彩色高分辨率的图像中提取出主要建筑物的位置信息并进行三维重建,提出一种结合2-D和3-D信息识别建筑物,通过纹理集技术进行大规模3-D重建的方法。进行边缘检测,提取可用的短直线以及相应的2-D特征,对这些短直线进行逐级聚类得到候选屋顶集合;通过朴素贝叶斯分类器在候选屋顶集合中区分出不同的3-D屋顶特征,以识别全局优秀的屋顶;根据屋顶的位置信息,在大规模的3-D场景中绘制出相应尺寸的模型,通过动态分配算法将建筑物纹理合并,减少纹理数量,从而减少纹理状态的切换。以航拍和卫星遥感彩色高分辨率图像进行实验,实验结果表明,该方法有优秀的识别正确率和3-D重建效果。
For the sake of recognizing roofs and 3-D rebuilding in high-resolution color images, a method of recognizing buildings based on 2-D and 3-D information and rebuilding the buildings based on the way of texture atlas was proposed. First of all, the image was processed using the edge detection, and lots of short line and 2-D characters were obtained. After a process of cluste- ring based on the classification system, the muster of potential roofs was gotten. At the last, naive Bayes classifier was used to get different 3-D characters and global optimal roofs. Based on the information of roof, the building model was drawn in 3-D scenes. And the textures were combined using dynamic distribution, reducing the number of textures and the texture state changes. The experimental results show that the method has good recognition rate and 3-D rebuilding nresentatinn
出处
《计算机工程与设计》
北大核心
2015年第1期191-196,226,共7页
Computer Engineering and Design
基金
国家863高技术研究发展计划基金项目(2012AA011801)
关键词
目标识别
边缘检测
霍夫曼变换
动态空间分配
渲染效率
target recognitiom edge detectiom Hough transformatiom dynamic distribute
efficiency of rendering
