Urban Building Detection And Reconstruction From Point Clouds Derived From UAV Oblique Imagery

As a new type of geospatial data,point clouds derived from UAV oblique imagery is playing an increasingly important role in the spatial data framework of Smart City.New methods of data processing and building extraction are required to meet the new demands in urban planning,monitoring and management.In the present research,we aim to process and extract buildings on two levels,namely building detection and reconstruction.This paper provides a new research perspective and research ideas for point clouds processing and buildings extraction.It also has a good demonstration effect on the large-scale integrated application of UAV image-derived point clouds.The main conclusions of this paper are listed as follows,(1)In the respect of building detection,we proposed a SVM vegetation detection algorithm and 3D morphological algorithm.The vegetation detection algorithm is not only beneficial to differentiate vegetation and building based on shape constraints,but also could greatly reduce the amount of data and processing difficulties.The morphological algorithm is help to smoothing the edge of the building,breaking the narrow intermittent between buildings and eliminating small protrusions,and reducing the influence of noise and mismatched data.(2)In the respect of building outline reconstruction,we proposed a line growing algorithm based on least square and W-k-means orientation algorithm.The line growing algorithm can solve the problem of contours extraction from edge points.The weighting algorithm enabled our solution to be robust to the data resolution,noise points and possible artfacts.(3)In the respect of roof reconstruction,we proposed a plane growing algorithm based on least squares and minimal circumscribed rectangle algorithm.The plane growing algorithm can solve the problem of planar patch extraction from roof points.The minimal circumscribed rectangle algorithm solves the problem of orientation and localization.