Lab 5-Construction of a point cloud data set, true orthomosaic, and digital surface model using Pix4D software

Introduction:

• What is Pix4D?

It is a photogrammetry software that can be used to generate point cloud structures and drone maps.  

• What Products does it generate?

Digital surface models, orthomosaics, and point cloud structures.

• Why is a so integral to UAS data processing,

This is integral to uas data processing because allows for quick and easy processing of the data and generation of maps.

Methods:

For this lab we used Pix4D, a software designed to generate an array of products from aerial imagery, to process the flight data that was given to us from our professor. This flight took place over a house, and we began the process by starting a new project within the software. We updated the camera settings as per our instructors request, and let Pix4D process the images. There were no GCPs used for this flight, so after the initial processing we confirmed the settings and began processing the orthomosaic. We also had the software create a DSM and a 3d cloud mesh so we could view the map in depth. The initial processing took 32 minutes and 12 seconds and used 66 out of the 67 images. 

• What is the overlap needed for Pix4D to process imagery?

Pix4D recommends at least 75% overlap and at least 60% side-lap.

• What if the user is flying over sand/snow, or uniform fields?

Pix4D recommends using at least 85% overlap and 70% side-lap. This is because it has hard to differentiate between points of an image with similar terrain like water or snow so the software has a harder time processing the images. A higher overlap will help solve this issue. 

• What is Rapid Check?

Rather than taking a longer amount of time to process the points normally this feature allows the user to save time by rapidly creating tie points faster. This may result in a lower quality final output, but the processing time will be less. 

• Can Pix4D process multiple flights? What does the pilot need to maintain if so?

The software can process multiple flights as long as the spacing between them is uniform with the same amount of overlap between the flights so the software can correctly generate the tie points. 

• Can Pix4D process oblique images? What type of data do you need if so?

For these images there needs to be ground control points for each set of images, and a high overlap so the software can differentiate between them.  

• What is the difference between a global and linear rolling shutter?

A linear shutter has a much longer capture time and does a scan of the image which can distort the image over the time it takes to capture it. This can lead to errors in data and DSM creation. Global shutter takes the picture in one quick motion and usually leads to less error. 

• Are GCPs necessary for Pix4D? When are they highly recommended?

They are not but come highly recommended because they can help correct error in images due to the characteristics of flight, and can especially be helpful for nadir images and oblique images.  

• What is the quality report?

The quality report provides a summary of all the work that was completed by the software. It shows the final output, how much overlap was detected, the number of tie points, how the blocks were adjusted, etc. The image below shows the overlap report from the quality report. 

Figure 1: Overlap

Results:

After the data was processed in Pix4D we used another program, ArcGIS to create aerial maps and digital surface models. The final Pix4d output can be seen below in figure 2. This orthophoto and digital surface model were both used later in ArcGIS to create the final maps. From this image you can see where some of the difficult points of processing are. For example, things like the trees and parts of the road did not come out as smooth as the areas with a more solid texture like this house. This is because the software has a difficult time differentiating various textures when there are very similar characteristics between them, which is why some parts of the mosaic did not come out well. This can lead to the blending of parts of the image, like around the tree for example, which is why the orthomosaic is not perfect. In other areas, like the ground around the road the map has a very high resolution, to the point where individual leaves can be seen clearly on the ground. 

Figure 2: Orthomosaic and DSM from Pix4D

Below is a table that shows the total processing time:

Orthomosaic     14:25

Digital Surface Model     0:48

Point Cloud        28:20

3D Textured Mesh          3:45

Total     67:00

Looking at the digital surface model above one can notice the variations in height that exist on the map. When looking at the image alone, it is very difficult to differentiate the various height changes that exist in the map. Obviously things such as trees and the house will have height to them, but when adding the hill-shade effect to create the digital surface model you can now observe the height change of the terrain as well. Note the red-ish color in the middle of the map, and the green outside edges of the map. This means that the area surrounding the map are much lower than the center of the map, and that the terrain has a steady increase in altitude as you travel more towards the center of the image. 

Below there is a video that shows a more in depth view of the orthomosaic. This is especially cool because Pix4D can generate a 3D point cloud view of the image, allowing the user to view the picture in 3D and essentially "fly" through it with a birds eye view. By using this feature one can get a better sense of how the final product was processed and appears because you can zoom in to various points of the 3D image and closely inspect it. On top of that, it provides a point of view that is closest to reality. Instead of just a top down image where it is hard to compare various heights and distances, you can literally "stand" in the image and look at it as if you were on the ground. In my opinion this is the coolest and most useful feature of the software because it is simulating real life. 

Video 1: Layout of Orthomosaic

In figures 3 and 4 below are the maps I created for this lab. These maps are helpful to the user because by adding a base-map beneath the orthomosaic and DSM it shows not only the accuracy of the flight, but also where the data was collected. As can be seen in figure 3, the GPS data that was used for this flight was not entirely accurate because the road on the base-map does not perfectly align with the road that was created in Pix4D from the actual images. This slight variation in coordinates is very common, and can be accounted for by a series of error points. An easy was to adjust for this deviation is to include ground control points, which the software can then use to cross reference the points from the flight and allow for corrections that would have the two maps align properly. However, for this lab we were not using GCPs so this deviation between the two maps is to be expected. Figure 4 has the DSM instead of the orthomosaic, and two side frames that show the user various points from the orthomosaic that the DSM if referencing. These maps both have all the proper elements to create an aerial map: title, north arrow, scale, reference window, basemap, metadata and watermark. 

Figure 3: First Map Creation

Figure 4: Second Map Creation

Conclusion:

Pix4D is very beneficial to the UAS world because of its ability to rapidly process UAS data in a user friendly way. Anyone is able to affordably acquire the software and quickly understand how to operate it, and easily output maps, digital surface models, 3D point clouds, and impressive animations that help simulate how the flight was flown for example. Even powerful tools such as this can have limitations and Pix4D is no exception. Issues can stem from deviations in data due to flight characteristics, and also create issues with areas of an image that are oblique or have very similarly colored points such as water or snow. It does also take a significant amount of time to process large data sets, so a power computer is recommended to process larger files. The data size for this lab was 67 images and took over an hour to process, and that time will rapidly increase with more images. Having a powerful computer with a RAM capabilities is highly recommended for processing UAS data.