Research On Detection Of City Building Change Based On Image Point Cloud

City change detection is of significance for not only geographic information updating and geography conditions monitoring,but also urban planning administration,urban dynamic monitoring and illegal building recognition etc.The traditional field survey is precise,however,inefficient,which is not available for these rapidly changing cities.In recent years,satellite image-based method has been adopted for illegal construction.However,due to the low level of automation of image processing,manual intervention is still in a high proportion.How to efficiently perform the building change detection based on satellite images has increased interests.With the development of laser scanning system,using three-dimensional information for building extraction and change detection becomes increasing popular.Due to the high cost of airborne light detection and ranging(LiDAR)and rapid development of three-dimensional reconstruction technology,aerial image matching point cloud-based data processing provide a new research perspective and idea.Taking aerial image data in two-phases as experimental data,dense point cloud and orthoimages of survey area were produced using SfM intensive matching algorithm and the field measurements were performed to calculate the precision and confirm its reliability.In our implementation,airborne LiDAR data processing method was first used to extract building information.The progressive triangulated network densification filter algorithm was adopted for filtering the ground points.Due to aerial image resolution,overlap degree and intensive matching algorithm,matching point cloud data quality is relatively lower than LiDAR point cloud,which results in the fact that geometrical characteristics of both LiDAR point cloud and dense matching point cloud are essentially different.Therefore,random forest classifier was used for point cloud classification based on spectral and vegetation indexes of image objects.Then,Euclidean cluster analysis and fa?ade removal were performed on the classified buildings.As a result,the roof points were extracted for building change detection.Based on the characteristics of orthoimage and dense matching point cloud and the analysis on building change attributes,a building change detection procedure was conducted.Using data organization structure of visual grid,buildings was grouped into four categories including unchanged,newly-built,demolished and rebuilt.In order to avoid registration error and classification error,pseudo-variation region was removed.Compared with the ground true,confusion matrix was statistically analyzed and its precision was evaluated.Moreover,the categories of building changes could be visual displayed on orthoimages.Therefore,the proposed building change detection method is convenient,practicable and precise,which has the vital practical significance.