我如何准确地使用GS18快速轻松测量？

TheLeica GS18 Iis a versatile and easy to use GNSS rover that uses Visual Positioning technology to measure points remotely in images. The system integrates a GNSS sensor with an IMU and a camera. Due to its precisest sensor fusion, it is possible to measure inaccessible points in images right away in the field. I explained inQ&A on Visual Positioning and Leica GS18 Ihow the GS18 I captures and processes images. With this expert insight, we will take it one step further. I will describe some fundamentals of photogrammetry and take a closer look at the automated matching process that allows measuring of survey-grade points in images inLeica Captivate.

How is it possible to measure points by selecting only one point in an image?

Immediately after capturing an image group, Captivate processes the GS18 I data and computes the position and orientation of each image. Therefore, the user can select one image, click on one point in it, press Measure and “瞧！” – 3D point coordinates have already been calculated in a global coordinate system. As you can see, the workflow of measuring points in images is effortless and straightforward. This is possible thanks to the highly precise and reliablepoint matching algorithmrunning on Captivate (often referred to as AR tracking).
这似乎相对简单。但是，您是否曾经问过自己的要点如何匹配？为了回答这个问题，我首先要解释一些摄影测量的基础。

Photogrammetry is the science of making measurements from images. The position of one point can be reconstructed from images that are positioned and oriented in a local coordinate system. The position of one object point can be defined by intersecting bundles of image rays, like in Figure 1.

Figure 1: Intersecting bundles of image rays

更具体地说，图像射线从相机的透视中心开始，穿过标记的图像点，并像图2中一样进入无穷大。

Figure 2: A perspective centre and an image ray

The object point we want to measure can be at any point along that image ray. To calculate the exact position of that point, at least two spatially separated image rays that intersect at one point are needed. These two rays must be defined by two different images. By increasing the number of image rays used for the reconstruction, the position accuracy will improve.

定义图像光线的方向,用户typically have to mark the point in each image manually. This is not needed when using images captured with the GS18 I. The following video nicely animates each step of the point matching algorithm, demonstrating how it automatically matches the marked point in the other captured images.

As shown in the animation, by marking one point in a selected image, the corresponding image ray will be computed. To define the direction of the second image ray, the same point has to be marked in the second image. The point matching algorithm does this automatically by connecting both perspective centres with a baseline. Now, using both the baseline and the first image ray, it is possible to create a plane. This plane is a so-calledepipolar plane, and it intersects the second image along the red line called theepipolar line.

对外相线对于匹配算法的点至关重要，因为第一个图像中选择的点位于第二张图像中沿着外两极线的某个位置。因此，该算法仅沿该行搜索最佳匹配。首先，Caintivate定义了模板矩阵，围绕着第一个图像的标记点的灰度像素的19 x 19矩阵。在动画中，模板矩阵由绿色构成。在第二张图像中，算法检测到点所在的外侧线段所在，并仅沿该段进行矩阵扫描。通过这样做，处理时间减少了。在扫描过程中，该算法沿着异性线的选定部分为每个点提取19 x 19像素矩阵。

在下一步中，该算法搜索最佳模板匹配。因此，将从第二个图像提取的每个矩阵与第一个图像的模板矩阵进行比较。这是通过计算矩阵之间的相关性来完成的。与模板具有最高相关性的提取矩阵被视为最佳匹配。然后使用该矩阵的周围像素以子像素精度找到点的确切位置。迷人的可视化与蓝色符号的匹配点可视化，并在所有匹配点的图像中出现。

匹配算法的观点有多聪明？

When developing the point matching algorithm, the aim was to create an algorithm that is as good at matching as the human visual sense is. However, it is clear that artificial and human intelligence cannot work in exactly the same way. For example, in many use cases, the point matching algorithm easily matches a point that could not be matched by the user. Look at the example in Figure 3.

Figure 3: Point marked in one image (left) and matched in another image (right)

On the left screen of Figure 3, one point on the pipeline is selected in the image. On the right screen, the same point is automatically matched in another image of the image group. One GS18 I user asked an excellent question: “How is it possible to automatically match the marked point in other images? Every point along this pipeline looks exactly the same to me, and I cannot see one unique point on this pipeline that I could manually match in two images. So, how can the algorithm do this if I cannot?”

答案很简单。正如我前面解释的那样，当一个图像中标记一个点时，匹配算法首先为每个图像创建一条外观线。然后，该算法沿二极分线进行搜索，以获取点的最佳匹配。如图4所示，外两极线与管道上的红线相交，在相交点，找到了最佳匹配。这就是算法很容易与人眼无法区分的两个图像中的点匹配的方式。

Figure 4: Epipolar line

Sensor fusion, photogrammetry and cross-functional development to solve surveyors’ problems

The Visual Positioning technology is using photogrammetric principles for remote point measurement. In addition, sensor fusion gives GS18 I the ability to join GNSS and IMU data together with the captured images. The unique combination of both photogrammetry and sensor fusion simplifies the traditional photogrammetric workflow. What is more, the point matching algorithm speeds up the measurement process, and it even helps users measure points that cannot be matched manually in images. This way, users can easily measure points in images with survey-grade accuracy. Not only is the mapping from the images possible onsite, but the same workflow also continues in the office withLeica Infinity.

In the GNSS team, we constantly push the boundaries to develop new solutions that solve surveyors’ problems. By developing a sensor that offers a simple solution to measuring challenging points, we want to extend the possibilities that surveyors have when measuring with a GNSS rover. With the GS18 I, we undoubtedly proved that even the greatest challenges could be mastered with synergistic teamwork. We did this for our users to have the ability to accurately and reliably perform remote measurements with survey-grade accuracy when using a GNSS rover.

To learn more about theLeica GS18 I, please visit:www.sendai-torema.com/GS18I