Research On Road Extraction From Remote Sensing Image And Point Cloud

The increasing ability of remote sensing data acquisition provides an increasingly abundant source of data for rapid updating of geographic information elements.As the “skeleton” of a city,roads are important elements of geographic information,which are closely related to urban development,transportation and battlefield environment analysis.Using remote sensing data to capture road information quickly and accurately has always been a hot research hotspots.The existing road extraction methods have great disparity with the application demand in road network extraction quality,extraction efficiency and automation degree.In this paper,the extraction technology of extracting road network based on high resolution remote sensing image,airborne LiDAR point cloud and fusion data source is deeply studied,and several improved ideas and methods are put forward,then the effectiveness of the improved method is verified through experiments.The main work and innovation are as follows:(1)A road extraction method based on improved K-means for high-resolution remote sensing image is proposed.Aiming at the problem that traditional K-means clustering algorithm is prone to noise interference and large amount of iterative computation,this paper proposes improvement measures.Firstly,the image is preprocessed,the road vehicle is removed by the multi-directional morphological opening and closing reconstruction method,the noise interference is removed by the guidance filtering method,and the contrast between road and non road is improved by contrast enhancement.To simplify the road network extraction model,the K-means clustering number K is set to 2(road class and non-road class);and GPU is used to accelerate parallel clustering algorithm to improve processing efficiency.Through a series of experiments,it is verified that the improved K-means algorithm has good noise immunity,and the extraction efficiency is improved significantly.(2)A road extraction method based on skew balance of airborne LiDAR point cloud is proposed.Aiming at the problems of large amount of computation and susceptibility to noise interference in the extraction of roads based on point clouds,a method of using the skewness balance algorithm to extract the on-board LiDAR point cloud roads is designed.Firstly,an efficient KD-tree index relationship is constructed for point cloud data to achieve fast neighborhood search.Secondly,the region growth method based on slope features is used to extract the road points,while the noise interference is avoided effectively and the search efficiency is improved.Finally,the road network is refined based on the spatial location characteristics of the road.Through the experimental analysis,it is proved that the method can extract the right road network from point cloud data,and has a high degree of automation.(3)A road extraction method combining high resolution remote sensing image and airborne LiDAR point cloud is proposed.The method is based on the characteristics of remote sensing image and point cloud data,and integrates two kinds of data to achieve the complementary advantages and extract high quality roads.Firstly,the initial road points in the LiDAR point cloud are extracted according to the skewness balance algorithm.Then,the SLIC super pixel segmentation method is used to segment the image,and the extracted initial road points and the segmented remote sensing images are fused to obtain the initial road network with high integrity.Finally,the use of local gray consistency detection to separate parking lots,squares,etc.,and then use the tensor voting algorithm to further refine the road and detect road intersections.By setting up multiple sets of comparison experiments,it is verified that the road quality of the fusion image and point cloud data extraction is higher than that of the road based on single data extraction,and it has high quality advantages in the same way of extracting roads by fusion data.The integrity rate can reach 93.59%,the accuracy rate can reach 89.45%,and the overall quality can reach 84.23%.