Accurate Positioning Methods Of Land Observation For UAV Oblique Images

Since the emergence of oblique photography in the last century,its application has been gradually expanded from military reconnaissance to the field of surveying and mapping,and it has become a hot technology in the beginning of this century.As the oblique images can provide more abundant information of features at lower cost,it is widely used in various fields such as urban 3D modeling,urban planning,urban security,government aided decision-making,military surveillance,and national defense,etc.Oblique photography has greatly expanded the application scope of traditional aerial remote sensing,and become a key link of data processing in the construction of digital city,smart city and smart earth,as well as the integration of space,sky and earth remote sensing data processing.The data processing of oblique images is one of the bases to realize these macroscopic applications.In order to provide high precision positioning results,oblique images and other auxiliary sensor observation data are needed to be highly integrated,so as to unify them in a ground coordinate system.The core technologies of oblique image positioning include image matching and mismatch removal,block adjustment,etc.,in which the former provides the inputs of the latter.Due to various factors,such as different viewpoints,ground textureless,a large number of mismatches are inevitably mixed in the automatic processing of these data.If there are no automatic mismatch removal methods,a large amount of manual interventions are necessary,further reducing the efficiency of data processing.In extreme cases,it will make subsequent block adjustment unable to obtain the optimal results,further affecting the positioning accuracy.In the data processing of oblique images for positioning,the imaging characteristics of oblique images,the singularity of normal equation,gross tolerance for solving the normal equation and other factors will affect the convergence and accuracy of block adjustment.Therefore,it is of great significance to propose corresponding solutions in block adjustment model.In some specific land observation tasks with remote and large oblique images,due to the cost and confidentiality of ground control conditons,the flexibility of tasks,the temporality of observation mission and the timeliness of data processing,the high precision control conditions cannot not be availuable,thus the traditional photogrammetric block adjustment may be difficult to provide high precision results for the reason that the observation geometry may be not ideal,further influencing the implementation of tasks.In this case,new constraint conditions should be sought to enhance the rigidity of block adjustment and reduce the dependences on ground control conditions(such as ground control points(GCPs)),so as to improve the accuracy.For example,laser rangefinder can provide the depth constraint information of observation targets or regions on some specific carrier platforms,and it is particularly important to realize highly fusion of observation data by taking advantage of the joint observation advantages of these multi-sensors.Therefore,this thesis takes unmanned aerial vehicle(UAV)as the observation platform to study the mismatch removal problem of matching points in the automatic data processing of oblique images and the new theory and method of block adjustment.The main contents of this thesis are as follows.(1)An adaptive mismatch removal method for oblique images,i.e.,AIMR is proposed.The hypothesis-verification methods(such as RANSAC,MLESAC)are often used to eliminate mismatches in the field of traditional photogrammetry.However,these methods usually require some sensitive parameter settings(e.g.,the distance threshold for point to the eplipolar line),while it is unable to determine the optimal threshold because of different scenarios such as structure,image distortion,viewpoint,which affects the subsequent automatic data processing.In order to avoid the sensitive parameters settings,this thesis utilizes the vector field interpolation theroy to solve the mismatch removal problem,and the adaptive method that resists to the sensitive parameters via the vector field interpolation framework is specifically designed.On basis of AIMR,considering the different imaging characteristics of viewpoints,this thesis proposes a mixture likelihood model of the anisotropic Gaussian and Uniform distributions that takes consideration of the resolution difference in x-and y-directions on an oblique images,then introduces it into the solution framework of AIMR,forming a more robust oblqiue image mismatch removal method——AOPM.This thesis makes a detailed experimental comparison and analysis of the two methods by making use of the UAV oblique images and the public datasets with viewpoint changes.The experimental results verified that these two methods are significantly better than the traditional RANSAC algorithm and MLESAC algorithm in terms of self-adaptability and efficiency,and slightly better than the latter in terms of the precision-recall tradeoffs.In addition,the experiments also confirmed that AOPM inherited the adaptive ability of AIMR,while the former had better ability of medium mismatch removal,so as to obtain a higher precison.(2)The original epipolar adjustment is improved,and an epipolar adjustment method with the constraints of additional ground control points is designed.The unknowns of the traditional photogrammetric block adjustment mainly include the exterior orientation elements of the images and the coordinates of ground points.Since the number of conjugated points is much more than the number of images,the memory consumption of block adjustment increases with the increase of the number of conjugated points.In order to reduce the storage and computation resources of block adjustment,this thesis regards the eplipolar cost as the main observation equations of block adjustment,uses the epipolar distance as the observations,combines with the constraints of additional ground control points,and builds a new epipolar adjustment method so as to unify the images to the coordinate system of ground control points.The unknowns of this method mainly includes the exterior orientation elements of the images and the coordinates of the ground control points,which greatly reduces the unknowns and thus reduces the memory requirements.The simulation experiments and real data experiments show that this method can converge to(or close to)the global optimal solution only with a small number of iterations.On basis of its results,the traditional block adjustment will converge quickly.(3)A polar coordinate block adjustment method based on weighted robust estimation and contraints of additional ground contral points is proposed.In the traditional photogrammetric block adjustment,the Euclidean coordinates of the tie-points are usually used as the parameterized expressions,howerver the normal equation is easy to be ill-conditioned when the base-height-ratio decreases,further affecting the convergence performance of block adjustment.On the basis of analyzing the different parameterized expressions of the tie-points,taking into account the convergence performance and the burdens of linearization,this thesis designs the corresponding block adjustment model,which is mainly based on the collinear equation expressed by the polar coordinates and supplemented by the collinear equation expressed by the Euclidean coordinates of ground control points.On this basis,the inconsistent vertical and horizontal resolution on the ground of oblique image points is also taken into consideration,and the corresponding weight design is carried out.At the same time,considering the stability of highly overlapped image points,a corresponding bounded loss function is introduced to suppress the influence of gross errors on the results of solving unkowns.In this thesis,the method is verified by multi-group UAV multi-camera oblique image datasets.The experimental results show that the proposed method has better convergence performance than the traditional method,and the accuracy is also improved to some extent,which verifies the effectiveness of the proposed method.(4)Combining with the laser obserbation data,a new laser ranging-aided block adjustment method of oblique images is designed.In terms of UAV joint observation system integrated with laser ranging and oblique images for local area or target tracking observation,its positioning accuracy is limited if there are no constraits of ground control points in the large oblique and remote observation environment and only the traditional GNSS/IMU-assisted bundle adjustment is used.Therefore,this thesis makes use of the depth constraint of laser ranging,and realizes the joint processing of laser ranging observation,oblique image observation and other data via the virtual conjugated points of laser ground footprint points on the image,so as to enhence the rigidity of the regional network under the situation where there are no ground control information,and finally improve the positioning accuracy.In this process,two modeling methods of laser ranging measurements are emphatically introduced,and the corresponding bundle adjustment method is also designed.Finally,the validity of the method is verified by simulation experiments.The experimental results show that the proposed method can achieve higher positioning accuracy than the traditional photogrammetric GNSS/IMU-assisted bundle adjustment under the condition of oblique remote observation.