Abstract: Semi-automated building extraction and reconstruction from aerial images are one important research topic of photogrammetry.This paper uses least-squares model-image fitting algorithm to extract and reconstruct rectangle buildings from aerial images.Firstly，the primitive model of rectangle building is given with parameters，then the parameterized model is projected to images according to the collinearity equation，then the least-squares model-image fitting algorithm is used to calculate exact parameters.Experiments are done to prove validity of the algorithm.

Key words: 3D building reconstruction, least-squares model-image fitting, rectangle building model

摘要： 针对航空影像中矩形建筑物半自动重建进行了研究。该方法基于最小二乘模型影像匹配算法，通过使航空影像中提取的建筑物边缘与模型投影线之间距离和最小计算建筑物的参数。给出用CSG与B-Rep相结合描述的矩形体基本模型，人工在影像上选点建立建筑物的初始模型，将初始模型反投到影像上，并对影像进行边缘提取，根据最小二乘模型影像匹配算法求解最佳模型参数。实验结果证明该方法有较高的准确性。该算法除航空影像与影像方位元素外不需要其他辅助数据，只需人工计算初始值，其他计算由计算机完成，速度较快，节约了人力物力，具有较强的应用价值。

关键词: 建筑物三维重建, 最小二乘模型影像匹配, 矩形建筑物模型