Automatic Mosaic Of Multi-source High Resolution Remote Sensing Images Assisted By LiDAR Point Cloud

Digital Orthophoto Map(DOM)is the image data obtained by digital differential correction of the original image included in DEM model.It is widely used in land surveying and mapping,resource monitoring and other fields.In order to get a large-scale DOM,and to reduce data redundancy,it is necessary to inlay multiple orthophoto images.The key step is the generation of the seamline.On the one hand,it reduces the manual work,on the other hand,the better seamline makes the final mosaic image have better quality.According to different tasks,the selection of the optimal seamline is different.For high-resolution remote sensing image,because the ground details are more abundant,such as cars,houses,vegetation and so on,the selection of the seamline needs to bypass the houses,high vegetation and other obstacles to prevent the image crossing the seamline.At the same time,the selection of the seamline needs to consider the image occlusion regions to prevent the seamline from bypassing the house on one image,but passing through the edge of the house on another image.In this paper,the occluded regions of the obstacles on the image called the occlusion regions.The automatic mosaic of the remote sensing image can be divided into three steps: first,building the topological relationship between multiple input orthophoto images,generating the initial seamline network;then optimizing the initial mosaic network;finally,synthesizing an orthoimage using the optimized mosaic network and the input orthophotos.This paper focuses on the first two steps.The traditional method without auxiliary data only considers the information of the image itself as the basis for the selection of seamline,so it is difficult for seamline to automatically avoid the obstacles and occlusion regions.Especially in the areas with dense houses,the direction of seamline is irregular,that is,in some regions,it bypasses the houses and passes through the houses in other regions.In this paper,we research on airborne Li DAR point cloud and multi-source high-resolution remote sensing image,design and generate an effective seamline generation method,and realizes the automatic mosaic of large-scale multi-source high-resolution remote sensing image.The main research contents of this paper include:1)This paper studies the algorithm of seamline network generation in the test area.This paper improves the algorithm of seamline network generation which is based on Voronoi diagram.A variety of post-processing methods are implemented,such as hole repair and nodes adjustment,to adapting to the characteristics of small and different overlapping ratio of multisource remote sensing images.2)A new automatic seamline generation method of two images assisted by Li DAR point cloud is proposed.This paper proposes a high-quality seamline generation algorithm based on trusted region assistance.The algorithm can search for an initial seamline on Li DAR point cloud,and then expand on the image with a certain radius to get the trusted area,reduce the optimization range of the seamline,which can improve the efficiency of the algorithm to a certain extent.At the same time,an algorithm of seamline optimization based on 3D terrain products is proposed.Firstly,DEM and DSM are obtained from Li DAR preprocessing to assist the optimization of dual image seamline.In the optimization of large-scale and large data amount satellite image seamline,the algorithm has the same efficiency as the method without auxiliary data,and has the same optimization effect as the method based on trusted area assistance.3)In this paper,an automatic generation algorithm of two images seamline based on 3D plane projection is proposed.In this method,the problem of projection difference is considered,that is,the geographical position of the same object is inconsistent in different images.The buildings and other obstacles are abstracted as an effective combination of multiple 3D planes.The 3D planes are extracted from Li DAR point cloud,and the projection geometry model is used to calculate the image occlusion regions to assist the seamline to automatically avoid them,3D plane extraction can be obtained and saved in the pre-processing of Li DAR point cloud,which can greatly improve the efficiency of the method and make the method real-time.