**5. Application research**

#### **5.1 Improve stereo images 3D positioning accuracy**

With the characteristics of small ground footprint, high sampling frequency, and large number of beams, the ground observation means of photon counting LiDAR represented by ICESat-2 has greatly improved the planimetric accuracy and flight direction data density of the acquired point cloud, which can be applied as a new 3D control condition to improve the positioning accuracy of satellite images [54–62]. To solve the application problem of photonic point cloud without synchronous image recording plane position, an approach to improve satellite image positioning accuracy with the support of satellite-based photon counting laser point cloud is proposed in this paper; firstly, a 3D terrain profile matching method is used to achieve accurate alignment between photonic point cloud data and DSM automatically generated from satellite stereo image; then terrain feature points are extracted from photonic profile point cloud based on slope change and combined with DSM multiple terrain features to generate common terrain feature control points, and finally introduced into the block adjustment of satellite images with attached parameters as the flat height control condition to further improve the positioning accuracy [63–66]. The experimental results using ZY-3 images and ATLAS ATL03 level data from two regions in Shaanxi Province show that the method can significantly improve the location accuracy without GCPs of satellite images, and compared with the fully uncontrolled positioning and SRTM data-assisted positioning methods, the planimetric and elevation positioning accuracy of ZY-3 image can be further improved by up to 60% and 34%, respectively, which verifies the effectiveness and feasibility of the method [67].

The test data were obtained using ZY-3 satellite images and ICESat-2 ATLAS data in the Xi'an area of Shaanxi Province, China. The ZY-3 satellite image was acquired on April 07, 2019, and the test field contains various types of terrain such as mountains and plains, with elevation relief up to 1200 m. The geographic location of the test area is shown in **Figure 11**, which is a stereoscopic (including front-view, normal-view, and back-view) image area ranging from E108.34°N34.36° to E109.02°N34.92°. The resolution of the front and back view images is 3.5 m, and the normal view image is 2.1 m. The 5-track ATLAS ATL03 data are used, and the acquisition time range is October 26, 2018–January 11, 2021, and the geodetic datum is WGS-84 coordinate system, the plane coordinates are latitude and longitude, and the elevation coordinates are WGS-84 ellipsoidal height, and the distribution on the image is shown in **Figure 11**.

Based on the dense matching using satellite stereo images, the minimum height difference method is used to realize the 3D terrain matching between the photon profiling point cloud and the generated DSM. The basic principle is shown in **Figure 12**, the dashed box is the search range, the solid red-green-blue line is the original three-beam photon point cloud ground track, and the dashed line is the distance search interval of the photon point cloud in its nominated accuracy range.

*Spaceborne LiDAR Surveying and Mapping DOI: http://dx.doi.org/10.5772/intechopen.108177*

#### **Figure 11.**

*ZY-3 stereo images and ICESat-2 data.*

*Diagram of 3D terrain profile matching between multi-beam photon point cloud and DSM generated by stereo images.*

The basic method is to first convert the DSM into regular grid data, take one of the tracks of the multibeam laser point cloud strip as the unit, move in the horizontal or vertical direction with a certain search step Δ*x* or Δ*y*, find the same point of the two data plane coordinates, if there is no corresponding point, use its close interpolation, and then calculate the absolute elevation difference between each laser point and the corresponding plane position of the DSM on multiple strips, until traversing all areas in the step setting area to determine the minimum absolute elevation difference position between corresponding points, which is the plane coordinate position of the profile point cloud matched with DSM.

The principle of terrain matching between profiling laser points and DSM based on the minimum elevation difference method is that the absolute elevation difference between the satellite photon point cloud and DSM in the local range is minimum, the accuracy of laser point cloud is high, and the accuracy of DSM is low, and the offset of the two data relative to the plane position is calculated by elevation constraint. The basic equation is shown as follows.

$$d\_{\min} = \text{MIN} \sum\_{t=1}^{3} \sum\_{l=-m}^{m} \sum\_{j=-n}^{n} \left( \sum\_{i=1}^{N} |Z\_i(\mathbf{X} \cdot \mathbf{Y}) - h\_i(\mathbf{X} + l\Delta \mathbf{x} \cdot \mathbf{Y} + j\Delta \mathbf{y})|\_{\mathcal{N}} \right) \tag{2}$$

In the formula, *s* is the number of beams, *d* min is the absolute value of the elevation difference between the photon point cloud strip and the DSM data generated from satellite images along the profile, *N* is the number of laser points in the strip, *Zi* is the laser point elevation value, and *hi* is the elevation value of the DSM generated from satellite images of the corresponding path. With this condition, the minimum elevation difference alignment can be constructed to obtain the best matching position of the photon profile point cloud strips relative to the direction of the satellite imagegenerated DSM trajectory, so as to obtain the overall offset of its spatial position.

The slope of the terrain is used as the basic discriminator, which is an important indicator to describe the terrain features and can indirectly reflect the relief pattern and structure of the terrain. For discrete photon point cloud data, a Gaussian fitting method is used to fit the curve, which can obtain more reliable terrain feature points. Then the terrain feature points are extracted from the elevation profile of the satelliteborne photon point cloud tracks based on the slope change, and all the terrain feature points with slope change values larger than the threshold are automatically labeled by setting the slope change threshold as the criterion. Take ICESat-2 point cloud data as an example, the extraction effect is shown in **Figure 13**. The continuous photon point cloud forms the terrain profile, and after the Gaussian curve fitting, it forms a smoother curve.

The principle of laser terrain feature point joint satellite image geolocation is to introduce the laser point as the control condition into the RFM compensation equation, to solve the problem that the parameters to be solved as free unknowns under the uncontrolled condition will lead to unstable accuracy of adjustment [68–75]. The basic process is to firstly match the connection points of satellite images and perform the

**Figure 13.** *Terrain feature points extracted from the photon profile point cloud.*

### *Spaceborne LiDAR Surveying and Mapping DOI: http://dx.doi.org/10.5772/intechopen.108177*

free network adjustment; then bring the joint topographic feature points as the planeheight control condition into the block adjustment of satellite images with additional parameters, and different weights be set to participate in the adjustment according to the observation accuracy of different data until the iterative calculation converges; finally, count the adjustment accuracy and output the adjustment results, i.e., the refined image positioning parameters, including RFM model coefficients and additional parameters, which are used together to realize the improvement of satellite image positioning accuracy. The characteristics of this adjustment scheme are that, due to the dense and continuous data of photon point cloud flight, it provides the possibility to automatically extract the topographic features common to both laser and stereo images with active and passive 3D observation data, thus providing an effective way to determine the position of photon point cloud on non-synchronous images, which enables its advantages of both plane and elevation accuracy to be fully exploited.

Comparing the results of uncontrolled, SRTM DEM-assisted and ICESat-2 control points, the uncontrolled block adjustment is based only on the self-contained rational function model, with a plane positioning accuracy of 8.12 m and an elevation accuracy of approximately 8.99 m. By introducing 5 m resolution open-source DOM and open 30 m grid spacing SRTM elevation data, and adding certain accuracy of plane and elevation geometric constraints, the accuracy of SRTM-assisted stereo image positioning is 6.65 m in plane and 2.019 m in elevation. When 1-track ICESat-2 laser point cloud control points are added to the test area, the 3D positioning accuracy of the image can be significantly improved to 2.64 m in plane and 1.39 m in elevation. Compared with the fully uncontrolled positioning and SRTM data-assisted positioning methods, the planimetric accuracy improved from 8.12 m, 6.65 m to 2.56 m, and the elevation accuracy improved from 8.99 m, 2.019 m to 1.319 m, with an improvement of 61.5% and 34.8% respectively compared with SRTM data. In the results from **Table 2**, we can see that the photonic laser point cloud can provide sufficient number of control point data, and the accuracy of auxiliary satellite stereo image positioning can reach the design theoretical accuracy.

The satellite-borne photonic point cloud has new breakthroughs in improving the positioning accuracy of satellite images compared with the existing linear system laser altimetry data and open-source global DEM: first, as a control of the adequacy of data distribution; second, the improvement of data accuracy, especially planimetric accuracy. This all provides a new technology path for achieving global accurate mapping.


#### **Table 2.**

*Accuracy statistics of uncontrolled block adjustment and auxiliary block adjustment.*
