A Photogrammetric Method for Spatial Data Extraction from Google Earth and Improvement with Precision Analysis

Topography maps are crucial for civil engineering projects, such as road construction, water channel construction, urban construction, and mining. Here we present a method which enables us to extract topographical map via modeling Google Earth and some field works. In this method, first, we model Google Earth as an object with closed-range photogrammetric method in the Agisoft Photoscan. Through some field works, we measured twenty-two points including twelve ground control points (GCP) and ten independent check points (ICP). Due to these GCPs, we were able to transform our model to real world with global polynomial and multi-quadratic equations and ICPs were used for precision analysis. This method is easy and cheap to obtain spatial data and the accuracy is sufficient for research requirements.


Introduction
Google Earth is used widely among researchers as a virtual globe which provides many free satellite images and three-dimensional views of Earth [1].The most prevalent use of Google Earth by researchers is to put their own collected data on the background images and observe the geographical context [2].The use of Google Earth is classified into some main objectives, namely visualization, data collecting, data exploration, data integration, modelling and simulation, validation and decision support by Goodchild and Stensgaard [3][4].The quality of images varies from place to place and the accuracy may not be suitable for some research purposes at certain locations in which case image processing is necessary for accurate data collection [5][6][7][8].Use of single coordinate system which is WGS84, easy visualization and accessible data, and freely accessible remotely sensed images are the advantages of Google Earth, but the inconsistency in the accuracy of its data is one of its disadvantages [1].A way to solve this problem is to integrate a database collected during field works and the data produced by the digital map which needs some further processing [9] and conversions [2,10].
In this article, a novel method is presented to extract data from google earth with higher accuracy.First, Google-Earth is remodeled using a photogrammetric method.Furthermore, transformation from model coordinate system to ground coordinate system is done by global polynomial and multi-quadric equations.Finally, the topography of a region is created in UTM coordinate system.The flowchart of research is shown in Figure 1.Google Earth is a free source of spatial data which has many users.Accuracy of Google Earth is different in various regions.One of its features is presenting a good visual space.Google Earth gives us an option to capture images with different resolution.Our images were taken with 4800×2718 resolution.Several images were taken from various views which overlap with each other like as modeling an object in closed-range photogrammetry.In order to image processing to build the model, several photogrammetric processing software exist.These include Australis, Photomod, PhotoModeler and Agisoft Photoscan.Since Agisoft Photoscan is a low cost software and also has high capabilities, we used it for our image processing purposes.
Agisoft Photoscan is able to detect conjugate points by performing automatic matching.The workflow of the software is presented as follow: after adding images to workspace, photos are aligned and corresponding points are determined; then dense cloud is constructed and depth filtering is applied; after that the discreet space is interpolated into a continuous space and a triangular mesh is built which forms a DSM; finally, a texture is built based on orthophoto mapping mode and mosaic blending mode.Although the algorithm used in Agisoft Photoscan is inaccessible, we are able to assume that it is based on bundle adjustment since adjustment of all images is done simultaneously in a single block of photogrammetry which means all the rays that go through the principle center of images are simultaneously adjusted.Basis of mathematical models for bundle block adjustment is the collinearity equation, given below: Where ,  are coordinates of image points stated in photogrammetry imagery coordination system;  0 ,  0 , − are interior orientation parameters;  0 ,  0 ,  0 , , ,  are exterior orientation parameters; , ,  are coordinates of ground points stated in ground coordination system, but GCPs are not used in the model space generated by Agisoft Photoscan and the transformation of model's arbitrary coordination system to ground coordination system is done by global polynomial and multi-quadric equations.The coordination system for our model is chosen arbitrary.The positional and elevation data of GCPs and ICPs are exported due to this coordination system.After model is built, we can export points, orthophoto, DEM, and also interior orientation parameters of camera.Since we didn't use a camera, interior orientation parameters were assumed free (self-calibration) in our equations.The interior orientation parameters of camera are as follow:  is the focal length;   ,   are principal point coordinates;  1 ,  2 ,  3 are radial distortion polynomial coefficients;  1 ,  2 are tangential distortion coefficients.

Field Work (Data Acquisition)
The considered region is placed in Iran, Kermanshah with 34°47.85′ − 34°45.1 ′  and 45°50.05′ − 45°51.3′.This region is slightly more than 3 2 .Figure 4 shows the location of the study.GCPs and ICPs are shown with red and yellow markers, respectively.The coordination of benchmark points was determined in Universal Transverse Mercator (UTM) coordinate system by dual frequency GPS.Benchmark points are used in order to enable total station to determine the coordinate system.The total station used is Leica TS09 and the coordinates of GCPs and ICPs were measured by it.The points are shown in Figure 3.The measured GCPs and ICPs are compared with Google-Earth software, and the corresponding root mean square is computed (Table 3).Due to the similar shifts to east in the points and the great value of RMSE, a systematic error exists.In order to resolve these errors, Global Polynomial and Multi-Quadric transformations will be used in next section.

Global Polynomial and Multi-Quadric Equations
In the presented method, mathematical statistics equations are used to transform Google Earth and our model to real world, in which we measured GCPs and ICPS.Global polynomial is a completely interpolative statistic equation that is used up to third order (cubic terms) for the transformation.The following equations represent third order Global Polynomial: Where , ,  are the transformed coordinates (GCPs and ICPs coordinates); , , ℎ are the coordinates of Google-Earth or model which should be transformed; {  ,   ,   ,  = 0, … , } are the coefficients calculated by GCPs.Note that (  ,   ) is a pair of terms for positional transformation and   is a single term for height transformation.
Generally, multi-quadric equations are developed form of point-wise equations.The amount of refinement and correction of these equations depends on different parameters of ground control points such as distribution, quantity and accuracy.After applying global polynomial, a deviation for , ,  could be calculated (, , ) and used in the following equations (multi-quadric equations): where   ,   ,   are coefficients;   is the distance from  th control point to the th control point in the horizontal plane;   is height difference of  th and  th control point.Therefore we could easily deduce that   = 0,   =   ,   = 0,   =   .From the above equations we are able to calculate the coefficients {  ,   ,   }.Then the deviation of unknown points could be calculated as follow: Where   is the distance between an unknown point and the th control point in the horizontal plane;   is the height difference of an unknown point from the th control point.Finally, the unknown point coordination is calculated:

Accuracy Assessment
In this section we will use ICPs to check the accuracy of the calculations.As we know, ICPs are dealt with like unknown points.By using GCPs, we derive the distance between GCPs and ICPs as follows: Where   = √  2 +   2 for positional, and   =   for height.

Results and Discussion
One of the outputs of Agisoft which was orthophoto has many applications as a photogrammetry output.This output could be used as data for remote sensing softwares such as ENVI, after being georeferenced.RS analysis results in cadastral mapping, vegetation maps and watershed mapping.Also it could lead to change detection of a region if orthophoto of different dates exists, and using Google Earth which contains dated photos in the presented method will make it possible.Figure 5 shows the orthophoto of modelled region.In our measurements, we had an elevation control point named "Killeh Sephid Valley".This control point was not considered due to the high difference of its measured value (556.43 m) and its value in Google Earth (590 m).Therefore, error of Google Earth in this point is high since it couldn't calculate the depth of Killeh Sephid Valley.6, error of the model is approximately equal to the error of Google-Earth.As it is seen in Figure 7, the best transformation to reduce horizontal RMSE is multi-quadric method with 9 pairs of terms.Model transformation with this method reduces the error with respect to measured values to 5.38 (m).The difference of this method with respect to Google-Earth is less than 1 meters.One of the outputs of Agisoft is generated DTM.Points exported from this DTM are transformed with the ideal transformation equation, and then used to create the georeferenced topography of the region in UTM coordinate system.

Conclusion
The method presented in this study could model the topology of a region with 5.3837 RMSE in horizontal and 0.14 RMSE in elevation.Since it was based on photogrammetric modeling of Google-Earth, the costs were low but the errors of Google-Earth affected our results.However, adding images from UAV or other satellite would improve the accuracy of the method.
From other advantages of this method we could mention being more supervised than Civil 3d and SketchUp where contours are automatically transferred from Google Earth to software.Also the accuracy is acceptable for studies and civil engineering projects with this much accuracy requirements.Places where have been changed due to natural/human causes and primary topography data is not available, our topography is the only reference which could be used as primary topography.

Figure 2 .
Figure 2. Capturing the area form various views

Figure 4 .
Figure 4.The location of GCPs and ICPs in the studied region

Figure 5 .
Figure 5. Orthophoto of the region modeled in Agisoft Photoscan

Figure 6 .Figure 7 .
Figure 6.Comparison between Elevation RMSE of Model and Google-Earth

Table 4 and
5 respectively, show the height and horizontal RMSE for global polynomial and multi-quadric transformations.Various polynomials have been checked and among them the best transformation which has the minimum RMSE are highlighted.

Table 4 ,
the most effective transformation in order to reduce RMSE is zero-order Multi-Quadric.RMSE of the transformed model and the measured values is 14 cm.