基于角度纹理特征和岔路口探测模型自动提取遥感影像道路网

Quasi-automatic Method for Extracting Road Network Based on Angular Texture Signature and Intersection Detection Model

针对已有的高分辨率遥感影像道路提取算法在自动化和连贯性方面的不足,提出了一种改进的道路网智能提取方法。使用传统的模板匹配等方法进行道路追踪提取时,无法识别岔路口,需要用户进行交互操作才能让追踪朝各支路方向继续进行。针对此问题,本文提出一种岔路口探测模型,用于识别岔路并通过获取到的岔路口信息自动指导各个支路的后续追踪。将该岔路模型有机结合到已有角度纹理特征匹配算法中,使相互连通的道路仅通过一次初始点设定便能全部追踪提取出来,极大地增强了提取的自动化程度。试验表明本算法是一种稳健的道路准自动提取算法。

This paper proposes a new enhanced road network extraction method, aiming to improve the shortcomings of traditional road tracking method in terms of automation and consistency. Traditional template matching methods cannot identify the junction or cross, and require extra user operation to initiate the tracking along each branch of the cross. To address this issue, this paper introduces a road intersection detection model, aiming to identify the intersection, extract the relevant information and automatically guide the follow-up tracking of each branch. Through combining this intersection detection model into the existing angular texture signature matching algorithm, the interconnected road network can be extracted only by one initial seeds setting, which significantly enhances the degree of automation. The experimental results show that the proposed algorithm is a robust quasi-automatic extraction method.