CLASSIFICATION OF TERRAIN USING LIDAR DATA

Abstract

This dissertation report describes the methodology adopted for classification of terrain using LiDAR data. With the availability of LiDAR technology (airborne or terrestrial) it has become possible to speedily capture the terrain information at high resolution with required accuracy. However, the raw LiDAR data cannot be used directly in the physical process modeling and there is need to generate a technique through which these data can be used for modeling purpose. One approach followed is to use LiDAR data to generate conventional data products which can be input to the data models. Buildings and other over-ground objects are separately classified and input into the models. The classification steps for different features work independent for each other. However, the error of omission and commission are not independent of the classes. So aim of the dissertation is to classify the terrain into three major classes a) ground b) buildings c) rest of other classes. First these classes are visually demarcated by using Terrascan software. After that same classes are classified by semi-automated procedures using Terrascan and error matrix is generated. The misclassified classes are carried into their true classes by use of echo, height, shape and intensity information. Misclassified data is compared with the first, second and third return of original data's surrounding point's characteristics such as return type, height, shape and intensity. If these points satisfy the surrounding point's characteristics then these misclassified data is carried to the contiguous class of the surrounding point. Therefore by using echo and height information misclassified classes points are to good level of accuracy classified.