摘要
针对传统的半全局匹配算法在处理视差变化大、遮挡严重的城市航空影像时,存在匹配精度下降、匹配效率低下的问题,提出了一种基于区域生长的半全局密集匹配方法。采用区域生长算法获取影像的初始视差,并从初始匹配点中挑选可靠的点作为视差控制点;利用区域生长获取的视差图,限制各个方向动态规划的过程以加速最优路径的搜索;通过视差控制点对动态规划的路径进行修正,避免错误匹配代价的传播。基于城区无人机影像的实验结果表明,所提算法不仅可以提高匹配结果正确率,还能使耗费的内存和时间都不到原算法的50%。
According to the fact of inaccuracy and less efficiency in Semi Global matching due to the large parallax changed and severe building block in Urban aerial images, this paper presents a algorithm of semi-global dense matching based on region growing. It gets the initial disparity map with the method of region growing and chooses some reliable points from the map as control points. Then, it accelerates the search of best route by limiting the process of dynamic programming with the initial disparity map. And it uses the control points to correct the route to avoid the spread of wrong matching cost. The experimental results based on city unmanned aerial vehicles image demonstrate that the algorithm in this paper not only improves the accuracy, but also reduce more than 50% of internal memory and time consuming compared with the original algorithm.
出处
《测绘科学》
CSCD
北大核心
2017年第5期12-16,24,共6页
Science of Surveying and Mapping
基金
江苏省自然科学基金项目(BK2012812)
关键词
半全局匹配
互信息
区域生长
动态规划
semi-global matching
mutual information
region growing
dynamic programming
