Building Change Detection Based On UAV Image Matching Point Clouds

The era of geographic information is coming quietly,which makes the development of digital city usher in a peak.Because of its low cost,convenience,automation and high efficiency of updating data,UAV can quickly and sensitively detect the rapidly developing areas of buildings and the rapidly changing areas of the city,which is suitable for the current urban planning.The UAV image is an ultra-high resolution image.Compared with the high resolution image of the satellite,the UAV image texture is clear and discontinuous,spectral segmentation is difficult,the same ground object is different spectrum,different spectrum of the same ground object,buildings and surrounding features can not be identified.It is difficult to obtain effective change detection results by using plane image information alone.UAV image stereo pairs are used for dense matching to form dense point clouds and stereo models.with the help of ground control points,elevation information can be obtained effectively and 3D data with high precision can be obtained.change analysis is carried out after iterative registration.The main contents are as follows:(1)According to the characteristics of UAV image processing,SIFT algorithm is used for feature matching,and then the matching image points with the same name are adjusted by the beam method with joint control points of POS end,and then the whole point cloud is obtained by encryption with PMVs algorithm.Finally,through the strict accuracy evaluation,it is judged that the precision plane of this dense matching is controlled in centimeter level,while the elevation is controlled in 1m,which fully meets the experimental requirements.(2)The improved ICP : divides the whole point cloud according to the appropriate density to construct the ICP initial matching feature,and uses the curvature and gray information as the matching feature to calculate the minimum Euclidean distance and accelerate the efficiency of point cloud matching.this algorithm does not need the calibration of a specific position,but only the feature information,so it can effectively match the two-phase point cloud and control it at the centimeter level.(3)Based on the complexity of the two-phase data registration,there is a big problem in the separation of ground features.This paper uses the CSF algorithm proposed in recent years,and the practice shows that the CSF algorithm can ensure the integrity of the building point cloud to the greatest extent,and the latter point cloud classification,point cloud segmentation and the separation of low buildings have the highest integrity,and the ground points can be separated as much as possible after complex point cloud registration.Building differential DSM,also has a good effect.(4)Sort out the interference factors of building extraction and change detection,and put forward countermeasures for each interference factor.The analysis shows that it is difficult to obtain ideal results by using aerial images or three-dimensional data,and tall buildings will cause shadow occlusion.Tall buildings and low buildings are effectively divided into two kinds of objects,and the idea of double threshold is used to detect the plane first and then the elevation change of high-rise buildings.For the remaining point clouds formed by low-rise buildings,the contour region is constructed by ray method,and the changing area is detected by integrating factors such as elevation,shape and area,and then the best region is obtained by voting on the Voronoi.Finally,the paper analyzes the detection accuracy and time efficiency of the processing results.