The Technology Of Aerial Triangulation Of Vehicle-borne Multi-camera Rig Images

With the development of application field of surveying and mapping industry,the sensors are placed on the vehicles,ships and animals besides aircrafts and satellites to obtain geospatial information of different scales and different kinds of areas.The imaging sensors have developed from single camera to multi-camera rigs to obtain images with larger field of view and higher resolution.With the development of smart cities,the mobile mapping system has become an important technology of collecting street level urban geospatial information.The multi-camera rigs equipped with mobile mapping system can collect high resolution images of both sides of road effectively.A large number of images are collected by the mobile mapping system and the main processing method of these images is projecting the images captured in one station to a sphere,a cylinder or a cube and stitching them into a panoramic image whose horizontal field of view is 360 degree.The panoramic images are treated as a good method of visualization browse.Some researchers conducted feature point matching and aerial triangulation using panoramic images whose field of view are 360 degrees to realize the geometrical positioning.The spherical ideal model and the spherical rigorous model are the main two imaging models of panoramic images.The extra errors of image points are introduced in the processing procedure of model projection and stitching of overlap areas within one station,which reduce the positioning accuracy of both the spherical ideal model and the spherical rigorous model.This dissertation treats the mono camera images as input data to match corresponding points and conduct aerial triangulation.The problems of matching corresponding points and deleting wrong matches of mono camera images are discussed.The extended collinearity equations are introduced into the aerial triangulation of multi-camera rig images.The aerial triangulation positioning accuracies of extended collinearity equations,general collinearity equations,spherical ideal model and spherical rigorous model are compared and analyzed.The influence of parameters of multi-camera rigs on image point matching and positioning accuracy of aerial triangulation are analyzed and discussed.The main research contents and conclusions in this dissertation are as follows:1.The feature points matching between mono camera images of multi-camera rigs.This dissertation matches feature points based on mono camera images of multi-camera rigs.The POS data captured by the mobile mapping system and the maximal depth of interesting objects are used to determine the overlap relationship between mono camera images,and the feature points are matched pair by pair,finally the wrong matches are eliminated by various constraints.The experimental results show that the rate of mismatches is lower than 2%.2.The matching of road feature points between mono camera images of multi-camera rigs.The initial position and orientation parameters of mono camera images and the fixed distance between projection centers and the road surface are used to calculate the homography matrix which is used to resample images to eliminate the scale and projection deformation and make the template matching correct.The feature points are matched pair by pair and the wrong matches are eliminated by epipolar constraint and multi-baseline constraint.The experimental results show that the rate of mismatches is lower than 4%.3.Conducting bundle adjustment of multi-camera rig images using extended collinearity equations as imaging models.The extended collinearity equations treat the image points in mono camera images as observations and treat the images captured in one station as one imaging unit whose horizontal field of view is 360 degrees,which avoid the extra image point errors caused by spherical model projection and overlap areas stitching.The experimental results show that the aerial triangulation positioning accuracy of extended collinearity equations is higher than that of spherical ideal model and spherical rigorous model.4.Analyzing the influence of the ratio of baseline length to depth and the depth on the number of matching points using real measured data.The experimental results show that the decrease speed of the number of matching points between image pairs changes from fast to slow with the increase of the ratio of baseline length to depth.There is a turning point of this decrease speed.The number of matching points increases at first,then decreases and finally approaches a constant with the increase of depth.There is a maximum number of matching points.This turning point of ratio and this depth corresponding to maximum number of matching points can be determined according to different scenes and reasonable baseline length and depth can be determined according to the hardware condition and the measurement requirements.5.Simulating the different distribution of matching points in multi-camera rig images by changing the number of mono cameras,the horizontal field of view and the vertical field of view of mono cameras,and analyzing the influence of this distribution on the positioning accuracy of aerial triangulation of multi-camera rig images.Simulating the changing of the distance between adjacent projection centers of mono cameras and the baseline length of adjacent stations and analyzing theirs influence on the aerial triangulation positioning accuracy of multi-camera rig images.The experimental results show that,when the matching points are evenly distributed in the field of view of 360 degrees,the multi-camera rig images can obtain a high positioning accuracy of aerial triangulation,even though the distribution of matching points in the vertical field of view and horizontal field of view of mono cameras is uneven.The distance between adjacent projection centers of mono cameras has no influence on the aerial triangulation positioning accuracy of extended collinearity equations,but this distance influences the aerial triangulation positioning accuracy of spherical model significantly.The number of corresponding image points and the intersection angles of terrain points decrease with the increasing of baseline length between adjacent stations,which result in the decreasing of the aerial triangulation positioning accuracy.