Processing Data in Pix4D (No GCPs)
Introduction
In this lab, we were introduced to Pix4D. Pix4D is an easy to use software that is currently the most popular structure from motion multi-view stereo software, or software used for constructing point clouds. In the previous labs, we used data created with the Pix4D software in order to build the maps with the ArcPro GIS software. This time, we will start processing the data with the Pix4D software package. We will then make the maps with the generated data in ArcPro GIS in a clean presentation. Before we get to that, the questions below provide an introduction and background that would be beneficial before continuing onto the project:
What is Pix4D? What Products does it generate?
- Pix4D is a unique suite of photogrammetry software used for drone mapping. It is currently the most popular structure from motion multi-view stereo software, or software used for constructing point clouds. It does this by utilizing GCPs and photogrammetry to create these point clouds and DSMs.
Why is it so integral to UAS data processing?
-
This software is so integral in processing UAS data, as it takes simple pictures taken for a UAS and produces DSMs (Digital Surface Models). These DSMs can then be used by Pix4D, or other software, to analyze and process the data collected.
- This software is so integral in processing UAS data, as it takes simple pictures taken for a UAS and produces DSMs (Digital Surface Models). These DSMs can then be used by Pix4D, or other software, to analyze and process the data collected.
Methods/Lab Assignment
The goal of this lab as mentioned in the introduction was to take pictures captured by a UAS and process the data. The pictures that will be used were provided to us from Dr. Hupy. The mission was once again of the Wolf Creek Paving Company’s property in Milwaukee, Wiscousin. This flight was flown on June 13th, 2017 by Peter Menet at an altitude of 70m. An example of some of the pictures captured can be seen below in Figure 1: Example of Pictures Taken. These pictures were taken using a Zenmuse X5 with rolling shutter. In the simplest terms, this means that the camera takes a picture every second which will then be used to construct the point cloud. It is important to point out that this project will not be working with ground control points (GCPs)
Figure 1: Example of Pictures Taken
Once again, before starting this project, proper file management is key. The same file structure was used as the previous labs that consist of three folders, a collection folder, processing folder, and analysis folder. If interested, here is a link to the previous lab talking about it.
To start, Pix4D was opened, and a new project was created. When doing this, the program prompts for the name of the project. The name of every project should follow a similar format that includes the data, site, platform/sensor, altitude, and corrections. For this project, it was named 061317_Wolf_X5_70m_noGCP.
Next, Pix4D will then ask you to select the images. For this project, we selected a smaller subset of pictures (21 images) of the original area used in other projects due to the processing time of the software. Once the photos are selected, it displays the image properties. A majority of the time, Pix4D will get all the settings right, but it is always best to double check and not trust the defaults. For this instance, a few items had to be changed. One of the items was changing the shutter from a global shutter to a rolling shutter, as this is what the Zenmuse X5 uses.
The output coordinate system has to then be selected. For this project, we used the WGS 84 UTM Zone 16. Next, the processing template has to then be selected. We chose the 3D Maps template. After that, the finish button was selected.
The project screen then appears and shows the flight and the image acquisition pattern. This will give you a preliminary indication of the quality. If any unwanted images are displayed (not in the same general area typically), it will be able to be seen on this screen. For this project, all the images were good.
It is important to test the quality of the images before spending all the time processing the data. To do this, only 1. Intial Processing should be checked (uncheck 2. Point Cloud and Mesh and 3. DSM, Orthomosaic and Index). After reviewing the processing options and unchecking steps 2 and 3, the initial processing can be done. Typically, this will take some time depending on the size of the project. This lab took around 2 minutes and 41 seconds as seen in Figure 2: Summary of Quality Report after Initial Processing, and the quality check can be seen in Figure 3: Quality Check.
Figure 2: Summary of Quality Report after Initial Processing
Figure 3: Quality Check
The quality report will provide the information on the quality of the data accuracy and collection. It is important to read this report thoroughly to ensure problems can be addressed before filly processing. If it looks good, the rest of the processing can be done by checking 2. Point Cloud and Mesh and 3. DSM, Orthomosaic and Index. Remember to uncheck 1. Initial Processing or it will repeat that step, wasting a lot of time. After this has processed, a screen similar to the one in Figure 4: Completed Processing will appear. The blue and green indicate the camera angles used to capture the pictures.
Figure 4: Completed Processing
The data has now been processed. Dr. Hupy recommended exploring it, so that is what was done. In Figure 5: Point Clouds with Cameras, the point cloud feature was turned on with the cameras left on, and Figure 6: Point Clouds without Cameras, the cameras were turned off. In Figure 7: 3D Model, the area was zoomed in to see the stockpiles of the aggregate.
Figure 5: Point Clouds with Cameras
Figure 6: Point Clouds without Cameras
Figure 7: 3D Model
The data turned out really well. The terrain can be seen clearly and differentiated between the different types of material. There were only a few voids in the point cloud, but I speculate this is probably due to the amount of overlap that occurred. This can typically be corrected by increasing the amount that the images look for overlaps within the images. Other areas that seem to have a few voids were the areas with the machinery such as the bulldozer. I am assuming this might be just the shape of the object however. Below are questions and answers that explains the results more:
How does Pix4D organize the output files/folders? Relate this to data management/naming.
- Pix4D organizes the output files in a similar way to how we are taught to organize and name the files and folders. It uses a general 3 folder system that contains all the data which makes for an easy management system. This can be seen below in Figure 8: Pix4D Folder Organization:
Figure 8: Pix4D Folder Organization
What were the processing times?
- The processing times were as follow in Figure 9: Processing Time:Figure 9: Processing Times
The next step involves creating a video animation trajectory in the rayCloud. This was done by clicking the New Animation icon on the Create section. This prompts the New Video Animation Trajectory wizard to open up. From here, I selected the User Recorded View, as I wanted to set the waypoints the video would follow. After this, I set the way points and recorded a series of point of views for three different videos that can be seen below in Video 1: Overview, Video 2: Overview 2, and Video 3: Detailed View. The settings for the detailed view video were adjusted to slow the camera angles down.
Video 1: Overview
Video 2: Overview 2
Video 3: Detailed View
Since the data has now been processed, the maps can now be created using ArcPro GIS. The required maps include an orthomosaic, DSM, and a comparison of this DSM and the one in the last lab (One without GCPs and one with GCPs). Following similar steps as in the last lab (link to see the previous lab), I started with an overview of the location that was processed in this lab. This can be seen in Figure 10: Overview of Processed Area. I wanted to compare this area with the overall area that was used in the last lab, so I stacked both orthomosaics that can be seen in Figure 11: Comparison of Area
Figure 10: Overview of Processed Area
Figure 11: Comparison of Area
I then created a DSM of the area processed in this assignment and changed the color scheme. This can be seen in Figure 12: DSM of Processed Area.
Figure 12: DSM of Processed Area
Below, in Figure 13: Comparison of DSMs, shows a quick comparison of the DSMs in the processed area in this lab versus the overall area that was processed. This will be used later in the overall map, but in this figure, the elevation values are clearly different because this lab uses no GCPs unlike the previous lab.
Figure 13: Comparison of DSMs
To satisfy the requirements of this lab, a map was created with the orthomasic and DSM of Wolf Creek. This can be seen in Figure 14: Orthomosaic and DSM of Wolf Creek.
Figure 14: Orthomosaic and DSM of Wolf Creek Map
Lastly, a map was created to present the comparison of the two DSMs. This can be seen in Figure 15: Comparison of DSM Map. The elevation differs for both, as the data processed in this lab does not use Ground Control Points (GCPs).
Figure 15: Comparison of DSM Map
Comments
Post a Comment