Extraction Of Building Facade Detail Based On DIM Point Cloud

With the advent of 5G era,the construction of digital city and smart city has begun to enter a stage of rapid development.Smart city are informational cities which based on new information technology such as Cloud Computing and Internet of Things,and its construction and development put forward higher requirement on the construction technology in the surveying,especially in building model refined.A complete and detailed three-dimensional model of urban building requires a further description of the building facade.Therefore,the research on facade model construction and facade detail information extraction is very important.The data sources for constructing urban building 3D model mainly include point cloud and image data,Obviously,point cloud data are more refined for model construction.At present,building model reconstruction algorithm based on point cloud data are mostly aimed at Li DAR point cloud,such as the modeling method that combines the building outline and DSM elevation information.However,model constructed by this method lacks facade texture information,so it is necessary to collect facade image data for improving the facade model,and it has not achieved three-dimensionality in essence,so it is difficult to meet the requirement of smart city construction.In order to give full play to the role of point cloud in facade model refinement,this paper studies the filtering classification and detail extraction method based on Dense Image Matching(DIM)point cloud which acquired by photogrammetry.The main work and innovation of paper are as follows:1.Explains the generation principle of DIM point cloud,focusing on analyzing the similarity and difference between sparse matching for relative orientation and dense matching for generating DIM point cloud in image matching technology.Analyze the characteristic of DIM point cloud and summarize its advantages in facade model refined compared with the Li DAR point cloud to do preparation work for DIM point cloud data processing experiment.2.Filtering of fa?ade DIM point cloud.First,combine the Li DAR point cloud filtering principle to analyze the research progress of DIM point cloud filtering,dividing the DIM point cloud filtering algorithm into two categories: one is mainly learning from Li DAR point cloud filtering algorithm to filtering the DIM Point cloud,another is mainly proposed based on the data characteristic for DIM point cloud.Then,analyze the spatial distribution characteristic of the fa?ade DIM point cloud and divide the DIM noisy point cloud into gross noise point cloud and redundant noise point cloud.The radius filtering algorithm is used to remove the gross noise point,and the filtering method based on mathematical model construction and Logistic regression algorithm are respectively used to remove the redundant noise point in plane facade and curved facade to obtain the precise facade DIM point,preparing for the subsequent experiment to extracting fa?ade detail.3.Building facade detail Extraction based on SVM algorithm.First,analyze the main principle of the SVM algorithm and explain the calculation method of its classification model construction,then according to the color difference feature of facade detail,select the DIM point cloud sample and base on the point cloud color information extract the sample feature vector to train the SVM algorithm classification model,and finally achieve the extraction to fa?ade detail.4.Innovation.Different from the perspective in traditional point cloud filtering algorithm which regards the point cloud information as three-dimensional spatial information,this paper regards the information which contained in the DIM point cloud as the feature of the learning object and extracts the feature vector by these features.Learning from the machine learning,this paper take the steps of selecting point cloud sample selection,training classification model,and classifying test data to filter and classify the fa?ade DIM point cloud to explore the application of machine learning in facade DIM point cloud processing.