For the past four years, I’ve been involved with the capture and delivery of data derived from drones. My experience has provided me with a good perspective on the quality of the data that can be derived from Unmanned Aerial Vehicles or UAV’s as well as the typical perceived expectations from my clients. This has been a very educational experience, but not to dissimilar from the early days using Global Navigation Satellite Systems or GNSS.
First and foremost, accuracy is not indicative of precision, nor is precision a guarantee of accuracy. A simple definition of accuracy is the nearness to the truth, (theoretically correct value, never actually known). A simple definition of precision is the measure of the consistency or degree of refinement of a group of measurements. In surveying the terms most often used to express accuracy would be the mean of a group of measurements and for precision, it would be variance and standard error (or standard deviation).
The graphic below is a good illustration of the difference between accuracy and precision.
On the right, although the bullseye was not hit, the shots are distributed evenly over the target though the shots deviated greatly from each other – the average was at the centre. On the left, the group of shots are all tightly grouped and did not deviate very much from the average, but were not centered on the bullseye.
If you took the bullseye as being the “truth” or “correct” value, then the average of right-hand target is very close to the bullseye – it’s accurate, but not very precise. The average of the left-hand shots is grouped very tightly together – very repeatable and therefore very precise. The average of the shots is precise but not close to the bullseye and therefore not accurate.
So how does this relate to UAV data? The precision of a measuring instrument is determined at the factory where it is manufactured. If you are utilizing a camera with an 8-megapixel sensor, you will not get the same resolution as a 20-megapixel sensor. The accuracy of your data though, is entirely dependent on the tools used (how precise it is) and how you use them (methodology).
The two questions we need to answer are: how do we obtain accurate results and how do we ensure precise measurements? First, let us look at what kind of data we are going to get. The basic deliverables from our UAV will be ortho-rectified aerial images and a point cloud where each point has coordinate values – northing, easting and elevation (X, Y, Z).
If we wish to ortho-rectify our images, the photogrammetric software must be able to “rectify” the tilt, angle, and coordinates of each image. The software will translate the north, east, elevation, omega, phi, kappa and finally the scale for the images. When the images are stitched together, we get our ortho-mosaic that is geo-referenced and scaled. From this digital product, our software will then produce a point cloud to coincide in this space.
To be accurate, then we need to ensure that the coordinate system the digital output is produced in matches the true coordinates on the ground. This will allow us to compare our data to data produced by other professionals and to allow our data to be shared by different entities that would utilize it.
If we use a proven precise technology , it will provide us with the confidence that our data is within a certain tolerance – the maximum statistical deviation from our contract standards. This will also give our data a set reliability when our client uses it to make decisions.
Regardless of whether we use an RTK enabled UAV or one that would require ground targets to orient the data during the processing stage, those base stations and/or targets need to be surveyed and related to the local control system. The methodology used in surveying the targets or base station will ensure the accuracy of the results. By using an appropriate surveying technology as well as reliable control points to tie into, we can ensure that the measurements made are precise.
In addition to the technology used to position our targets or base stations, we also need to have targets distributed throughout the site. The location of our targets will enable the processing software to obtain a good fit. This will allow the calculation of the data to be both accurate and precise.
In our example below, I’ve illustrated how the targets were distributed on this project.
Once we have processed our data sets, we need to do a little quality control and quality assurance (QA/QC) prior to delivery. To do this, we collect additional data in the field that will not be used in the processing of the data set – additional ground targets, topographic ground measurements and perhaps a tie-in to a visible feature that can be measured accurately and precisely.
We tend to use the centerline of a roadway, a corner of a concrete pad and ground shots in a 3 x 3 array spaced approximately 3 metres apart. Upon completion of processing the data set, CAD software is used to place the image and created Digital Elevation Model (DEM) into the project. The next step is to import our QA/QC data consisting of ground shots, centerline and photo identifiable target measurements. A comparison of the measured elevations to the elevations derived from the DEM will be reviewed in conjunction with the statistical accuracy report from the photogrammetric software .
It is important to note that the QA/QC data sets should be derived independently of the photogrammetric process, but tied directly to the ground control to ensure the data is similarly geo-referenced.
In my next article, I will discuss targets – types, sizes and materials required to ensure visibility, durability and efficiency of placement.