Automatic Extraction Of Urban Road Network By Integrated Processing Of Aerial Images And LIDAR Data

Accurate and detailed road models play a crucial role in a number of geospatial applications,such as infrastructure planning,traffic monitoring,map updating,and driver assistance systems.In this research,we present an integrated method for automatic road extraction using combined data(high-resolution aerial images and LIDAR point clouds).In order to update the available maps,LIDAR data and aerial images are a promising data source for creating roadmaps to maintain activities and missions of government agencies and consumers.Despite the fact that large amounts of high-resolution aerial images and dense LiDAR data are being collected,saved and remaining unprocessed or unused,new data sets are continuously being accumulated.This problem is due to the fact that the evolution of automatic techniques for processing aerial imagery and LiDAR data is more limited as compared to the hardware sensor technologies.Object extraction for full exploitation of these data sources is very challenging?Remote sensing techniques such as measures from aerial imagery and LiDAR(Light Detection and Ranging)provide one means by which large areas can be rapidly mapped with a high standard of accuracy,but the technology that uses this sensor data for detection and mapping of road networks is still in its infancy.Although many different methods have been developed for the semi-automatic or automatic extraction of road information,the efficiency of a given method depends on image resolution and the road input characteristics,and also on the algorithms used(developed to extract the desired information,using a conmbination of appropriate image-processing techniques).Hence all the problems are still unsolved and none of the techn,iques can solve them without human interactions.Since there are big variations of roads(urban,rural,abrupt)in different cases of their complexities(trees,building,shadow,occlusion caused by different objects etc.),it should be noted that the existing algorithms for road generation and their functions are all task-based and data-based.However,the proposed approach identifies road ribbons and contextual targets(i.e.,Parking lots,grasslands and trees).Because the Lidar data contains a lot of information about the scenes,most of the terrestrial features such as roads and buildings are discenable.Furthermore,in LIDAR intensity images the road characteristics are homogeneous and have the same height as a bare surface in elevation and can,therefore,be taken as a mask to remove the effects of shadows and trees.In addition,normalized DSM(nDSM)obtained from LiDAR is employed to filter out other above-ground objects,such as buildings and vehicles.After that grasslands and tree areas have been detected and extracted from aerial imagery using color-based segmentation.Therefore,the proposed method generally focuses on improving the detection of the road network,whereas the precise delineation of roads has been less attended to.The accuracy of the results is evaluated by comparing the manually extracted reference data.Moreover,several quality measures(Completeness,Correctness,Quality etc.)are applied for accuracy assessment.The results show that this approach can extract more than 90%roads.