提出一种新的图像匹配技术解决增强现实(AR)系统中的户外3维注册问题。首先用数学形态学提取视频影像中的地形边缘曲线。根据GIS的通视分析原理，从DEM（数字高程模型）栅格图像上生成基于观测视点的地表视域轮廓。再分别检测两个轮廓线上的局部极值点，计算各自的极值点曲率向量。最终通过用相关矩阵度量两个极值点曲率向量之间的相似性获得匹配结果。研究结论表明，与地图相比，基于视频AR的地理信息可视化对地理数据建模的要求较低，且无须对空间数据进行抽象符号化。

A new approach of image matching is introduced for 3D registration of augmented reality system. First, mathematical morphology is used to extract the edge of video images. And the land surface visible portions of DEM are derived at the viewpoint based on GIS visibility analysis. Second, the local extremes in the corresponding contours are detected and their curvature sets are calculated, respectively. Third, the similarity of the corresponding curvature sets is calculated by comparing the similarity matrix.