Creating Orthomosaic in ArcPro

Introduction

The creation of orthomosaic is one of the most common ways to process UAS imagery data. In this lab, we used close to 200 images to create an orthomosaic. But generally speaking, what is an orthomosaic? This will be addressed. Also, professor Hupy showed us how tiepoints work, and how some of them were rejected for accuracy issues.


What is photogrammetry?
Photogrammetry is the subject, the science and an indirect measurement tool for photographs and digital imagery. The end product is typically 3D-models of objects and orthoimage mosaics (ortho maps). The production of orthoimage mosaics and digital elevation data is created through the concept of photogrammetry.


What types of distortion does remotely sensed imagery have in its raw form?
What is orthorectification? What does it accomplish?
Ortho mapping tends to focus on aerial photogrammetry, which covers areas from a distance and is typically nadir viewing (straight down viewing). To illustrate this, these two figures will display Professor Hupy's house from different angles.

Figure 1: Nadir viewing.

Figure 2: View from 60 degrees angle. 

In our case, we used a series of nearly 200 images that have collected as the sensor (UAV) has flown along a flight path. Typically, most images wear a distortion and therefore needs to go through a process of geometrically correct every single one of the orthoimages. The resulting images have the geometric integrity of a map. So, orthorectification is a method for geometrically correct distortions and produce accurate orthoimages. There is another concept as well. The product of stitching a number of orthoimages together into one layer is called orthomosaic. For every image, the sensor's position (derived from GPS, namely X₀, Y₀, Z₀) and orientation (the angles of w, j, k) is key. This displays where the UAV was at the very specific time of the image were taken. Other important features in this process are the tie points between the images and a Digital Elevation Model, preferably a high-resolution one.

If not the known ground positions (GCPs, x, and y) are delivered to the end-user, along with the images, the images need to be adjusted. This adjustment georeferencing process is done accordingly with the information one finds in the metadata.

Figure 3: Before the Block Adjustment

Also, taking elevation data into account generates a more accurate output but is more time-consuming. If the elevation data is unknown, it can be derived through stereo imagery from the overlapping images. When the Block adjustment is calculated the area and overlapping images were looking like Figure 4 below is illustrating. It looks distorted but is actually more accurate than Figure 3.

Figure 4: After Block Adjustment

Normally speaking about distortions, one refers to geometric and radiometric distortions. These two is removed during the orthorectification process. But there can also be lens distortion parameters on needs to take care of. So when creating an orthoimage from raw imagery one needs to pay attention to sensor distortions (so-called interior orientation).

With LiDAR images it is easier to find the ground since with laser one can return a number of echoes. Basically one can create both a Digital Terrain Model (DTM) a Digital Surface Model (DSM) directly with LiDAR technique. Since the level of accuracy increases the need for adjustment is reduced.

What is Bundle Block Adjustment (BBA)?
Bundle block adjustment is a computational process for fitting images together while statistically minimizing errors with the ground control. In order to compute the exterior orientation for each image, the process is using both GCPs and tie point information. The orientation for a whole bunch (called a block) of images is then adjusted to fit the ground.


What is the advantage of using this method? Is it perfect?

The obvious advantage with BBA is that the quality fitting images together increases when adjusting the residual errors. One of the disadvantages is that bloopers often occur in the BBA process. When trying to fit blocks of images to the ground together, there is a high likelihood the fitting is not performed perfectly. The anomaly points with high residual errors are often deleted and thereafter the adjustments can be recomputed until acceptable errors for each point is reached.

What is the Ortho Mapping Suite in ArcPro? How does it relate to UAS imagery?
The ortho mapping focuses primarily on aerial photogrammetry products and feature extraction software. Esri's tools are in the realm of UAS used primarily to produce different orthorectified products.

Methods/Lab Assignment

In order to create an orthomosaic, one has to go to the Ortho Mapping tab and under the Product group of features and click Orthomosaic. From there one follows the wizard, decisions about overlap (should be no less than 60%), balance method (preferable dodging), target raster (the reference base map data), and as the last step one can choose format. The Cloud raster format is a faster choice than TIFF selection, but latter is a lossless one and preserves all the graphical features of the image. After that, it is possible to start the process of orthorectification. The image below displays the processing time, 7 minutes and 52 seconds. 

Figure 5: Time elapsed for the orthorectification process.

What key characteristics should go into folder and file naming conventions?
File names should convey the most important information and what data the files contain.
Examples would be dates, names of locations, what kind of om digital model is used and file extension is key for naming files. Also, keep the filename without spaces. For instance: "20170613_wolfpaving_dsm.tif"

Why is file management so key in working with UAS data?
Working with UAS data means a lot of file types and extensions used. Good file organization is heavily aligned with the effectiveness of the user and it also helps other individuals to know what they are looking for.

What key forms of metadata should be associated with every UAS mission?
Examples of UAS metadata that could be included in every mission; the date of data collection, pilot name, weather conditions, UAS platform, sensor type, altitude, Ground Control coordinates, and UAS coordinates.

The key metadata we are working with, presented in the table in Figure 6.

Figure 6: Table with metadata for this UAS mission.

Results

My maps show Professor Hupy's home and wherein the state of Indiana the area of studies was performed.

Figure 7: The Bundle block adjusted UAS imagery and its location

Figure 8: After the block adjustment, this is the final compilation of stitching orthoimages together.

The quality in the center of the map, right below where the UAS data is retrieved, is superior. This is pretty comprehensible since there are more tiepoints that overlap each other there.

Figure 9: Many tiepoints and a big amount over overlap generate great quality of the map.

Figure 10 displays the outer layers of the image projection (the footprints), where the quality is poor and one can even see some clear break between the adjustments and orthorectification has failed.

Figure 10: Areas on the map where the data quality is poor

But still much better than the image from the reference data, the World Imagery.

Figure 11: Dr. Hupy's house displayed as World Imagery.

The imagery results can be seen in Figure 12 and the calculated statistics in Figure 13.

Figure 12: A transparent view of the elevation model. Layer included to the left.

Figure 13: Generated Statistics for the DSM

The creation of a Digital Elevation Model took about 45 minutes in total, seen in Figure 14.

Figure 14: In ArcGIS Pro, Digital Elevation Model also took some time.

Conclusions

To summarize the Orthomosaic Tool, one has to start with generating an ortho mapping workspace. Then a Bundle Block Adjustment (BBA) have to be performed before the establishment of ortho mapping products.

Creating the orthomosaic was not hard at all and took about 8 minutes. What was time intensive was the step before when we made the Bundle Block Adjustment, which took more than 30 minutes. We processed this through an online server. However, it would have been a lot faster if we would have done the process directly from c:\ instead. Another drawback, for some reason it was not possible to work with the GCPs at this stage. The quality of the process gets neater the more tiepoints and the higher degree of overlaps the images have.

 Using BBA is fine when accuracy is not a concern. For instance, using BBA when processing images taken from low altitude, generates sensible accuracy and a fast result. This wouldn't be recommended if the demand for accuracy (like in mining) was high nor when working in areas with erratic terrain, mountains, and valleys. The data created with UAS is more accurate than using images from satellites. But LiDAR is even more accurate than UAS imagery, goes down to center- or even millimeter.