摘要
剖面匹配在道路提取中得到了广泛的应用,但现有的相关文献中普遍利用数学模型模拟道路轨迹、利用最小二乘剖面匹配的结果进一步修正数学模型预测的道路点的方法,这种方法与实际的路况有较大的差别。本文中,用户初始化后获得道路的起始点、前进方向、道路的宽度和模板剖面。通过将模板剖面沿道路前进方向平移和旋转生成一系列目标剖面,求目标剖面与模板剖面的灰度差的平方和,在求平方和时给剖面的道路部分更多权重,最小的平方和对应的目标剖面的中点为最精确的道路点,迭代上述步骤追踪道路轨迹。经试验证明,改进的剖面匹配算子是一种稳健高效的道路追踪算法。
Profile matching is widely employed to trace road. Currently, most of the relevant algorithms build a mathematical model to fit the trajectory of road and to predict the position of road centerline points, then get the precise position through the least square matching between model profile and search profile generated at the predicted point. But it is not the case that a mathematical model can＇t fit all geometric shapes of roads in practice. So an improved method of the profile matching is presented in this paper. The start point, its moving direction, and the width of the road are given by initialization, a model profile is generated, then we shift and rotate the model profile to generate new search profiles, and then calculate the least squares between the search profile and model profile, giving the road part more weights in computation of the least squares, the smallest one corresponds to the optimal search profile whose midpoint is the precise road centerline point. Above steps are iterated to track road trajectory. It is proved that the weighted rotating profile matching is a robust and fast road tracker.
出处
《测绘科学》
CSCD
北大核心
2009年第4期64-66,126,共4页
Science of Surveying and Mapping
基金
国家973计划资助(高性能的遥感影像网络计算与信息解译的智能方法(20006CB701303)
国家863计划资助(高性能遥感数据集群处理关键技术(2007AA12Z151)
关键词
半自动道路提取
高分辨率遥感影像
剖面匹配
模板匹配
主动轮廓模型
semi-automatic road extraction
high-resolution remotely sensed imagery
profile matching
template matching
snakes
作者简介
林祥国（1981-），男，山东泰安人，武汉大学博士研究生，现从事遥感图像处理与信息提取方面研究。E—mail：linxiangguo@gmail．com
