When it comes to geospatial data analysis and mapping, accuracy is crucial. One important tool that helps achieve high accuracy is ground control points (GCPs).
In this blog post, I want to share my experience processing one of our recent datasets using different numbers of GCPs ranging from 1 to 5.
Join me as we explore the impact of these reference points on geospatial accuracy.
Table of Contents
What are GCPs and why do they matter?
Imagine you’re looking at satellite imagery or aerial photographs. These images can have distortions due to various factors like camera lenses or the shape of the Earth’s surface. The location information generated by the GPS module might also be a few meter off, particularly in the vertical (Z) direction. The accuracy is influenced by factors such as equipment quality and satellite geometry.
Therefore, people introduced ground control points to act as reference markers to align these images with the real world. By accurately determining the coordinates of these points using base station, GPS receivers, GCP markers (sometime you can also use a RTK drone instead), we can correct distortions and obtain precise geospatial information.
Working out how many GCPs are required
Before diving into the testing process, let’s take a moment to understand our setup. The dataset consisted of 107 images collected at an altitude of 80m using a DJI Phantom 4 Pro drone. During the data collection, we set up a base station and receiver, and we also placed five ground control point (GCP) markers at specific locations indicated in the picture below. These markers were positioned at each corner (GCP 2,3,4,5) and one in the center (GCP 1) of the surveyed area. Once the data is collected, the GCP information is formatted according to the requirements of the photogrammetry software. Then, I processed the orthomosaics and digital elevation models (DEMs, including DSM and DTM) using 0-5 GCPs respectively.
There are plenty sources available telling you why at least five GCPs are recommended. The reason behind is quite simple. I would always imagine this as stretching a bedsheet. You can’t really stretch it nicely if less than three points are secured on the mattress. However, map is not exactly like a piece of cloth. What exactly might happened if only one or two GCP is provided? Therefore, the testing begins.
GCP testing results
Now, let’s dig into the exciting part—the results! As expected, the geospatial accuracy of the dataset improved with the increasing number of ground control points. Images that had only one GCP showed noticeable distortions, while those with five GCPs exhibited significantly enhanced alignment and accuracy. But what exactly happened with our testing dataset?
Let’s first have a look at the result without using any GCPs. You can clearly see that the GCPs are not directly laying on top of the on-ground markers in the orthomosaic. There are about 8-9m difference between them consistently.
GCP marker location when orthomosaic is processed without GCP data provided
However the quality of the orthomosaic looks fine. Feature detection and reconstruction of all 107 images yield reasonable result compared to the satellite imagery. The height of the DTM and DSM ranged from 21 to 58m.
Satellite
Orthomosaic with no GCP
Left: Orthomosaic with no GCP provided VS Right: Satellite image
1 GCP
When only one GCP was used, the result was interesting. We selected the center GCP (GCP 1) and the top left GCP (GCP 3).
It’s like you ask someone to pin two pieces of paper together without any other instructions. The other person would probably not know where to pin and how to position two pieces of paper together. This happened to the map as well. It was only centred at the location we provided GCP information. The distance between the GCP and marker is less than 1 meter (still not perfect), yet the relative size of the orthomosaic is completely off.
With all 5 GCPs
With 1 centre GCP
Left: 1 centre GCP vs Right: With all 5 GCPs
With 1 corner GCP
With 1 centre GCP
Left: 1 centre GCP vs Right: 1 corner GCP
2 GCPs
When two GCPs were used. The result is a lot better. As you can see directly, most of the GCP markers are overlaying with the GCP location.
GCP marker location when orthomosaic is processed with 2 GCPs data provided
With all 5 GCPs
With 2 GCPs
Left: 2 GCPs vs Right: With all 5 GCPs
The different that can be observed from the result is that the orthomosaic seems twisted or rotated from the no GCP output and full five GCP output. This time it feels like the orthomosaic is a piece of paper that glue on the stick, so that it rotates along the axis, where two GCPs actually introduced some distortion towards the edge.
3-4 GCPs
In these scenarios, three or four GCPs gives similar result in terms of the overlay between GCP markers and GCP locations, yet map distortion still exists. We tested using GCPs located in a line (GCP 1,3 and 5), as a triangle(GCP 1, 2 and 5), outer corners (GCP 2, 3, 4 and 5), and clustered together (GCP 1, 2, 3 and 4). The transformation tend to lean towards the direction where the GCPs were provided.
For example, Comparing with the orthomosaic using all five GCPs, the orthomosaic with only three GCPs covered less area in the bottom left corner where GCP is missing.
With all 5 GCPs
3 GCPs (1,3,5)
Left: 3 GCPs(1,3,5) in a line vs Right: With all 5 GCPs
With all 5 GCPs
3 GCPs (1,2,5)
Left: 3 GCPs(1,2,5) in triangular position vs Right: With all 5 GCPs
GCP marker location when orthomosaic is processed with 3 GCPs (1,2,5) data provided
GCP marker location when orthomosaic is processed with 3 GCPs (1,3,5) data provided
GCP marker location when orthomosaic is processed with 4 GCPs (1,2,3,4) data provided
GCP marker location when orthomosaic is processed with 4 GCPs (2,3,4,5) data provided
5 and more GCPs
With all five GCPs provided, you can clearly see that all markers are correlated with the GCP location quite well. However visually not too much difference than the ones with three or four GCPs.
GCP marker location when orthomosaic is processed with 5 GCPs data provided
However, if you compare the orthomosaic without GCP input, there is a clear shift of the result.
With all 5 GCPs
With no GCP
Left: no GCP vs Right: With all 5 GCPs
Visualizations and examples above demonstrated the impact of GCP numbers on georeferencing quality, leaving no doubt about their importance. The results emphasize the significance of ground control points in achieving accurate geospatial analysis. However, it’s important to acknowledge that certain limitations exist. Factors such as image complexity, availability of GCPs, and the overall quality of the dataset can influence the final accuracy.
Ground control points (GCPs) may not be needed in scenarios where high absolute accuracy is not a requirement, and relative measurements or visual representations suffice. GCPs may not be necessary when the dataset is used for visualization, preliminary analysis, or non-precision applications. Additionally, collecting GCPs can be impractical or challenging in remote or inaccessible locations, or when limited resources or time constraints are present, making it more feasible to proceed without GCPs.
In conclusion, GCPs are the hidden gems of mapping, have the power to transform maps captured by
affordable consumer drones into top-notch survey-grade products. It is clear that increasing the number of GCPs leads to improved alignment and precision. As geospatial analysis continues to evolve, it is essential to consider optimal GCP quantities for reliable results.
Thank you for joining me on this exploration of geospatial accuracy and the impact of ground control points. If you’re interested in diving deeper into this topic, feel free to reach out to us and have a discussion. Happy mapping!
