Processing Pix4D Imagery with GCPs

Part 1 (Introduction):

Ground Control Points (GCP) are important in the processing of UAS data. GCPs allow for better orthorectification in that the help better align the data on the coordinate plane. Ground control points are points on the ground that have a known geographic location. GCPs can be used either as a GCP, or as a checkpoint. For a Ground Control Point, the rectification software will use the point to adjust the image, whereas with a checkpoint, the software will just give an error measurement from that checkpoint. This allows the user to see how accurately the data is being displayed or processed.

Part 2 (Methods):

Similar to the previous lab where we processed a set of data without the use of GCPs, this lab incorporated GCPs for processing the data. We first did initial processing of the data without the GCPs. Once the initial processing was complete, we added the GCPs.

To add the GCPs, we first told Pix4D which file was contained the GCPs. Note that the points had already been cleaned up, and the points that were to be used were the only ones contained in the file. Once this data had been imported, we instructed Pix4D as to which field contained the X value, which the Y value, and which the elevation value. Figure 18 below shows the GCP manager in Pix4D, and as you can see, it has detected each point as well as displayed the respective coordinate data.

Figure 18: GCP Manager

After the GCPs were added, we had to manually adjust the data to match the GCPs, as the plane from the UAS was not entirely accurate in the vertical direction. Figure 19 shows the sidebar where each GCP was adjusted, and Pix4D was told where on each image the control point was actually located.

Once this was complete, we told Pix4D to reoptimize, and it adjusted the data to incorporate the GCPs.

Figure 19: GCP Adjustment

Figure 20 below shows the difference between where the GCP was told to be placed, and where the GCP was actually adjusted to. This difference between the two points will be the error that was mentioned with checkpoints above, although this is GCP error.

Figure 20: GCP Difference

Once these steps were complete, we exported the data and created a few maps with the data. 

Part 3 (Discussion):

The first map below is the orthomosaic with the GCPs overlayed on top of the orthomosaic. One thing I noticed, that may be hard to tell as it is displayed, is how close each GCP is to its respective chevron on the map. This tells me that Pix4D was able to adjust the data to the GCPs fairly accurately, which I will examine more using the report.

Figure 21: Ortho Map with GCPs

Figure 22: DSM Map using GCPs

Attached below is the quality report for this data. If you scroll down to the quality check section on the first page, the georeferencing section shows a mean RMS error of 0.045m. This means that in 3D space, the average error of the points was 4.5 centimeters. If you scroll down to the bottom of the fifth page, you can see the error of each individual point. The highest error displayed is point 102 on the Z-axis, which shows an error of 0.142, or 14.2 centimeters. This is roughly 6 inches error in the vertical direction, which does not seem like very much for this data set and this large of an imaging area.

Quality Report.pdf

Part 4 (Conclusion):

It is clear that GCPs can have a large advantage to the accuracy of the data, and one place we see this is with the Z-axis and the vertical elevation. Without ground control, the data would have been processed as if it was on average 26 meters lower than it actually was, and this was due to the elevation data from the UAS. With ground control, however, that data was corrected to the actual elevation, and we have much more accurate data because of it. Not only that, but on average, the data was 6.94 meters off in the horizontal plane as well (Using Pythagorean's theorem, the X direction error and the Y direction error were used to calculate overall horizontal error), so we can see the overall benefit of GCPs to improve the accuracy of our data.