Some thoughts on accuracy
50 cm resolution
40 cm resolution
20 cm resolution
10 cm resolution
5 cm resolution
Images courtesy of Scott McTavish
|
There are two crucial elements to
understanding the accuracy of your data and your math model.
The first is most obvious. You need to understand how accurate your “ingredients” for your math model are. If you measure the coordinates of a corner of a parking lot with your smartphone as a GCP, you are looking at an accuracy of several meters. Survey the same point with a high accuracy differential or RTK survey, and the GCP will be accurate to decimeter or centimeter accuracy. If you have a laboratory calibration of your camera and lens that gives you accurate distortion parameters, you’re going to have a more accurate model of the geometry creating your digital image than if you just have chip size and focal length. Using a block math model instead of individual images, linking your images via tie points, can help to improve the overall accuracy. But the old saying, “garbage in, garbage out”, holds. The second element is often overlooked, but perhaps even more of a consideration in the UAV market. Just how accurate do you need to be? With the increase in resolution and fine centimeter-level GSDs, it is tempting to strive for pixel-level accuracy in ground coordinates. But is this realistic? What are you trying to compare it to? If your GIS data layers are only accurate to a meter, then having a more accurate image layer isn’t going to make it overlay better. And obtaining pixel-level accuracy in a UAV is going to require more investment in quality optics, calibration, high accuracy ground surveys, increased manual editing, etc. If a quick and economical solution was the goal of the UAV in the first place, perhaps accuracy of a few meters is adequate. Are you looking to update precisely surveyed features, or perform metrics on stockpiles? Then consider the following items;
Are you looking to map erosion features on a stream bank to add to a lower resolution GIS layer? Perhaps meter-level accuracy with GCPs from your handheld GPS is plenty. Are you looking for a quick on-the-fly map of a remote disaster scene with limited ground access? Perhaps you have to rely on the on-board GPS and tie points to give you approximate ground coordinates. One of the persistent problems with accuracy crops up when overlaying new data over old and they just don’t line up. It is tempting to assume that there is a problem with your current project. Keep in mind that your UAV data may actually be more accurate, due to higher resolution and accurate GCPs, than your GIS layer that may have been digitized from a paper map. In such cases, you may need to decide whether you always want new data to be as accurate as possible, or whether to use the base data as your ground control source so that any new data will overlay properly. |
