Research On Building Detection In Urban Area Based On Aerial LiDAR Data And High Resolution Remotely Sensed Imagery

As important symbols of the city ground, buildings are vital to urban planning, hazard forecast, resource exploration, digital city and military reconnaissance, etc. Therefore, researches on the automatic buildings detection are of practical significance. The airborne LiDAR obtains 3-d information rapidly, accurately and directly, and the researches on the buildings detection from this data source have become the new development trend, but, without high overall accuracy. High-resolution remote sensing images contain rich spectral information, which is good compensation to the airborne LiDAR, and the combination will be beneficial to the automatic and accurate extraction of buildings.Based on the previous researches, this study presented a hierarchical classification algorithm that combines the airborne LiDAR and the high-solution remote sensing image to detect the urban buildings. Firstly, the normalized digital surface model (nDSM) is gained from the difference operation between the digital surface model (DSM) and the digital elevation model (DEM) after filtering. DSM is generated from the interpolated original LiDAR point cloud data. Based on the previous recognition of the research area, we select the minimum building height as the threshold value to filter part of the ground objects. Secondly, we apply the region growing algorithm to divide the nDSM after the elevation filtering into non-overlapping areas. Lastly, the vegetation confusingly similar to buildings is filtered along with two steps, namely, NDVI based and spatial relation based. The major contents and conclusions are as follow:(1) Organization of the airborne LiDAR data: through the analysis of the characters of the airborne LiDAR data and the quality of various data organization forms, we think that the regular grid can simplify the data organization form, reduce the algorithm hardness and increase the data processing efficiency, which is suitable for the airborne lidar data organization forms in urban areas that are relatively sloping gently. On the basis of this, we select the appropriate grid size and interpolating functions, determine the weights according to the distance between the laser foot point and interpolation point, and raise a new DSM interpolation algorithm. The result reveals that the DSM generated from this interpolation algorithm can effectively express discontinuities and prevent the mean variation of the traditional interpolation algorithm at the building edges.(2) Filtering of the airborne lidar data: based on summarizing of the filtering principle of the airborne lidar data, concluding the difficulties in filtering and synthetically analyzing and evaluating the advantages and shortages of the traditional filtering method, we put forward an improved filtering algorithm based on the plane fitting. This method, firstly, applies the region growing algorithm to carry out the image segmentation to the resampled LiDAR data, extract the maximum connected region as the primary ground, refine the primary ground points based on the sloping threshold value, and execute the filtering to the kriging interpolation of the selected ground points. The experiment applies the standard filtering experiment data provided by ISPRS. As shown in the results, compared with the segmentation-based filtering method raised by Brovelli, the improved filtering algorithm will reduce the error rate I from 21.75% to 1.47%, error rateâ…¡from 2.39% to 1.85%, and the overall error rate from 12.92% to 1.65%. The adjusted filtering algorithm will better solve the problems including the gross error influence, selection of original ground points and filtering of large-scale buildings, etc. It will effectively differentiate the ground points and the ground objects, which is more suitable for the filtering of the airborne LiDAR data in the urban area.(3) Classification of the airborne LiDAR data:based on the researches on the organization and filtering of the airborne lidar data, we study the classification of the data. As for the vegetation that is easily confused with buildings, this study will filter from two aspects. Firstly, we carry out overlay analysis to the vegetation extracted from the remote sensing images, based on the normalized difference vegetation index (NDVI), and the segmented nDSM images, filter the vegetation existing in the remote sensing images according to the overlaying area between them. Secondly, as for the hidden vegetation, we filter it according to the spatial relation between the buildings and vegetation and its size. As the result reveals, this study takes the relief displacement in the remote sensing images into consideration and solve the problem of direct registration between the airborne LiDAR data and the remote sensing images. Compared with the method that directly filter the vegetation by NDVI, this study increases the detection rate of buildings from 85.94% to 90.20%, and greatly reduces the FN error. Meantime, the research also shows that the extraction of the building edges from the airborne lidar data has lower accuracy than the extraction of the main building bodies.(4) This paper selects the study area in Gulou District of Nanjing, carries out the experiment on the airborne LiDAR data processing system developed by the author, and compare it with the data proved in the field. As the research result shows, the building detection rate of is 90.20% with the accuracy of 92.58% and the overall accuracy of 94.31%. It let us know that the method raised in this study possesses sufficient feasibility and relatively high utility value, which will effectively realize the automatic extraction of the 3-d information of the urban buildings. The future work mainly includes the improvement of the practicability of the filtering algorithm and the solution of the merge after the image segmentation, etc.