3D model reconstruction using aerial photo to measure object height

Abstract

3D reconstruction, especially for height extraction, using digital aerial photos taken from a non-metric camera and Unmanned Aerial Vehicle (UAV) is a challenging study. The purposes of the study are to determine the factors affect the accuracy of an object’s height reveal from a 3D model and establish procedures to deliver the optimal result. Two or more overlapping aerial photos can be constructed into the 3D model by applying the principles of collinearity and epipolar geometry using 3D reconstruction algorithm. Due to the instability of non-metric digital camera, the camera must be calibrated before 3D reconstruction is processed, in that way the quality of spatial extraction can be then measured. The study is conducted using 24 megapixels resolution Sony NEX7 digital camera and Hexacopter UAV. UAV was flown to a height of 50 m, 60 m and 70 m and placed at a certain position in order to get some height variations of camera. Camera Calibration Toolbox is utilized to calculate intrinsik parameters of the camera and a specific program is developed using MATLAB in order to build the 3D model and to obtain the object’s height. The result validation is done by comparing the height from 3D model with that one measured using Electronic Total Station. The accuracy of the object’s height up to 1 mm successfully achieved, with the largest height prediction error reaches 15.2 cm at 70 m flying height above ground level and average percentage error 4.07 for 20 objects.