Processing UAS Imagery with Pix4D

Background
Having already worked with data acquired from Pix4D, this lab served to introduce the software and walk through the processing of data using Pix4D. In the process, a mosaic would be generated of the Litchfield Mine in Eau Claire, Wisconsin. The use of UAS as a form of data collection has increased in recent years, and Pix4D is one of the leading software programs for processing this data. Pix4D works by analyzing multiple aerial photographs of an area for similarities and keypoints. Once these points have been located, the software combines the images into 3D cloud points and mosaic rasters. However, there are several critical points that must be discussed before utilizing the software.

• What is the overlap needed for Pix4D to process Imagery?
• The overlap required to process imagery is 75% frontal overlap and a minimum of 60% sidelap. High overlap is critical to getting accurate results. Thus, data acquisition must be planned to maximize overlap (Figure 1).

  

  Figure 1: An ideal flight plan for maximizing overlap.

   
• What if the user is flying over sand, snow, or uniform fields?
• When a survey area is comprised of large, invariable areas, it becomes more difficult for the software to properly match images. When flying over sand, snow, and fields, it becomes necessary to increase the overlap to compensate. In these areas, the minimum requirements are 85% frontal overlap and 70% sidelap.
•  What is Rapid Check?
• Rapid Check is a feature that allows for the quick verification to see if the flight settings and parameters were formatted correctly and will result in the creation of a raster. It does this by reducing the image resolution to 1 megapixel to decrease process time. If the Rapid Check fails, it is recommended that the flight be redone over the survey area. If the Rapid Check succeeds, full processing may commence. It is critical to return to full processing, as reducing the resolution to one megapixel results in a decrease in positional accuracy, which may negatively affect results.
• Can Pix4D process multiple flights? What does the pilot need to maintain if so?
• Pix4D can indeed process multiple flights. However, several criteria must be met.
• The flight plan must collect enough overlap for each image.
• The flight plan must collect enough overlap between the two or more images (Figure 2).
• The flights must be flown at the same altitude, in the same or similar conditions (sunlight, weather, etc.), and within a close enough time frame that the surface features have not changed.

Figure 2: A depiction of two flights with enough overlap for Pix4D (left), and a depiction of two flights without enough overlap for Pix4D analysis (right).

• Can Pix4D process oblique images? What type of data do you need if so?
• Yes it can. In order to do so, there needs to be multiple flights at multiple camera angles. One above the object or site, one at a 45 degree angle, and one at a 90 degree angle
• Are GCPs necessary for Pix4D? When are they highly recommended?
• GCPs are not necessary for Pix4d. However, having them does increase the positional accuracy and georeferencing. GCPs should be used when accuracy is of the utmost concern. This is typically done for city and street construction, corridor mapping, or other such urban construction plan where accuracy is critical
• What is the quality report?
• A quality report is a summary file that is created after the initial processing is finished. From the quality report, a variety of information can be gathered a preview of the image mosaic and a number of initial processing details.

Methods
First, a new project was created in Pix4D. It was titled "20160621_litch_krismejr_phantom3_60m" based on the date of the survey, the site, the sensor, the altitude, and the project creator. Then the images from the flight over the Litchfield Mine were added to the project. The camera model settings were edited so the Shutter Model read as Linear Rolling Shutter. All other camera settings were left as default. The output coordinate system was left as default, the processing template was set as 3D Maps, and the project setup was finished. The DSM, Orthomosaic, and Index processing options were changed to triangulation in the Raster DSM option. The initial processing was started and completed, with a quality report being generated afterwards. According to the summary all 68 of the images were used, with none of them being rejected (Figure 3).

Figure 3: The summary taken from the quality report after the initial processing. All 68 of the images take of the Litchfield were used in the initial processing, with none of them being rejected.

Also taken from the quality report, the majority of the image overlap occurs in the center, with the areas of low overlap being around the edges (Figure 4).

Figure 4:Number of overlapping images computed for each pixel of the orthomosaic. Red and yellow areas
indicate low overlap for which poor results may be generated. Green areas indicate an overlap of over 5 images
for very pixel. Good quality results will be generated as long as the number of keypoint matches is also
sufficient for these areas.

This makes sense, as the aerial sensor only rarely passes by the edge and flies over the center of the survey area multiple times. Afterward reviewing the quality report, the "Point Cloud & Mesh" and "DMS, Orthomosaic and Index" final steps were initiated. When this was completed, a triangle mesh. Using the built in functionality of Pix4D, a digital flyover of the mesh was created and exported as a video file. Finally, the raster files generated in the final step were brought into ArcMap and used to create cartographic ally pleasing maps of the Litchfield Mine, projected in the WGS 1984 UTM Zone 15N coordinate system and the Transverse Mercator projected.
Results

Figure 4: A raster imagery map depicted the Litchfield Mine in Eau Claire Wisconsin (right) with a reference map (left). 

Figure 5: A raster elevation map depicted the Litchfield Mine in Eau Claire Wisconsin (right) with a reference map (left).

Based on comparisons made between the imagery mao (Figure 4) and the elevation map (Figure 5), several inferences can be made. The highest elevation areas are mounds of soil and dirt built up over the course of the mine's operation. The flat areas are reserved for the mined areas, which have been reduced to relatively featureless areas of exposed earth. The lowest elevation areas are the forested areas which ring the mine site.

Conclusion

Pix4D has proven to be a available tool for a remote sensing analyst. Even without the use of GCPs, Pix4D was able to collect multiple photographs taken by a UAS over the Litchfield Mine to both create a 3D triangular mesh and a series of raster images which could be converted into maps. In the future, Pix4D could be utilized to construct something similar over a larger area with multiple UAS flights, or create a highly detailed and spatially accurate map with the use of GCPs in order to aide to city construction.

Sources

Hupy, J. (2017). Construction of a point cloud data set, true orthomosaic, and digital surface model using Pix4D software. Eau Claire, Wisconsin.