Processing Data with Pix4D

Date: 10/17/2024

Data organization

            Before we can get into the processing, the first step is to organize your dataset. The images from this dataset were imported off of OneDrive into the collection folder in the provided template for processing (Figure 3). This allows for easy access to the dataset when we go to import the images into Pix4D.

figure 3: File Template

Import the Dataset

            After creating a new project in Pix4D, select the processing folder within the template as the desired storage location for the finished product and name the project accordingly. For these missions, I named the projects “kwittme_091324_Skydio3DCapture” and “kwittme_091324_Skydio2DCapture”(figure 4). Providing metadata such as name, date, and type of flight performed helps with organization of projects and differentiating project files in the future. Pix4D will prompt the user to add images. At this point select all the images in the dataset using the ctrl+A function (figure 5). Pix4D will then display the image properties window which will display all the images along with the geolocation data.

figure 4: Project file names

figure 5: Selected images in the dataset

Image and Coordinate System Settings

While in the image properties window, it is important to enter the camera model editing settings and change the shutter mode from global to linear rolling shutter, as this is the shutter mode used by the Skydio S2 platform.  After confirming the image settings, Pix4D will give the option to modify the output coordinate system. For this project there was no need to edit the output coordinate system, however this may not apply for all projects as there may not always be a known coordinate system associated with the images.

Processing Options.

            Within the processing options template window there are a number of options available for templates, each of which applies to a different sensor or type of scan. For both projects we will select the 3D maps option (figure 6). Although this is misleading because one scan is technically called a “2D” Capture, it is still generating a 3D product. The mission is merely flown on a 2D plane and uses gimbal angles to create this 3D product. All other settings are set to default for these scans.

figure 6: Select 3D Models

3D Capture Results

            The subject of the 3D capture was a red trailer parked at the Turf Farm (figure 7). With a dataset of 207 images, the 3D capture took about 2 hours to process and generated a relatively high-quality image. The shape of the trailer was well formed with sharp edges. Additionally, there was a small trailer with a piece of machinery parked next to the red trailer. We were unsure of whether it would process well due to its shape and size, however it appeared very clearly in the image. The biggest defects in the product were effects of direct sunlight. The trailer has a polished aluminum roof which reflected light causing some distortion on the depth of the surface (figure 9). Also, the north side of the trailer displayed some surface distortion caused by shade and possibly lack of ambient light reflecting off the surface (figure 8).

figure 7: 2D Capture

figure 8: Distortion caused by shade

figure 9: Distortion caused by reflecting sunlight

2D Capture Results

            In comparison to the 3D capture, the 2D capture yielded much poorer results (figure 10). This project consisted of only 66 images and processed in about half an hour. All the issues caused by sunlight shown in the 3D capture were exaggerated in the 2D capture (figure 11). Additionally, the 2D capture did not display edges as sharp as the 3D capture, and blended the bottoms of both trailers into the ground (Figure 12). Adjusting the settings to increase the number of pictures taken such as increasing the overlap or sidelap may help improve quality, but ultimately the 3D scan likely provided a better product because it captured images at varying altitudes allowing it to capture more detail of the sides of the trailer. Some benefits of using 2D capture however would be the smaller size dataset, and therefore shorter processing time. Well lesser in quality, it is important to recognize that the 2D capture did produce a viable scan in a quarter of the processing time of the 3D capture.

figure 10: 2D Capture

figure 11: Distortion caused by shade

   figure 12: Edges blended into ground