Research On Filtering And Classification Algorithms Of LIDAR Data

Airborne lidar(light Detection and ranging)is one of the high-tech developing rapidly in recent years,it can quickly obtain 3D structure information on the ground,and is widely used in 3D mapping,vegetation protection,power line inspection and other fields.Point cloud classification is also known as point cloud semantic segmentation,point cloud semantic classification.With the development of airborne radar technology in recent years,point cloud classification has become a popular research direction.Contrast and analyze three kinds of point cloud filtering algorithms,which are filtering algorithm based on digital morphology,filtering algorithm based on irregular triangulation,and filtering algorithm based on cloth simulation.The ISPRS Vaihingen dataset of lidar remote sensing point cloud data is selected,and three algorithms are used to perform filtering experiments and analyze the filtering results.The accuracy of the filtering results based on irregular triangulation is lower than that of the other two algorithms.The overall filtering accuracy based on digital morphology is higher,but the type Ⅱ error is greater than that based on cloth simulation.Therefore,the filter results of cloth simulation are selected for point cloud classification.This paper proposes a method based on cloth simulation filtering and point cloud classification of random forests.The effect of terrain change on the accuracy of point cloud classification can be reduced by normalizing elevation value by the ground point cloud generated by cloth simulation filtering.Compared with the unnormalized results,the classification accuracy was improved by 30%.The fusion of point cloud information with remote sensing spectral information is further improved.For other classification results there is a 3%to 5%increase.