U-Net边界检测的实景模型单体化包围盒创建方法OA北大核心

Aiming at the problems of low automation and poor efficiency of traditional manual methods and com-plex algorithm and poor accuracy of point-cloud-filtering method in the process of feature extraction and bounding box creation,a dynamic singling bounding box creation method based on U-Net boundary extraction in true or-thophoto is proposed.Firstly,the U-Net neural network is used to extract the contour boundary of the true or-thophoto,and extracts the forward projection boundary of rural buildings efficiently.Furthermore,based on the extracted building boundary,the vector model of single building bounding box is automatically created by geo-graphic coordinate transformation and building height extraction through image dense matching point cloud.Tak-ing two villages in Jiangxi Province as an example,the experimental results show that the proposed method has high accuracy and efficiency in building boundary extraction,and the location and height of the bounding box cre-ated are accurate.It can effectively perform the dynamic singling of the real-scene model in large-scale buildings community.