Understanding Stitching

Behind the scenes of the map stitching process in drone photogrammetry

DroneDeploy
Jun 30, 2015 · 5 min read

By Ian Smith, Sales & Marketing @DroneDeploy

Read part 1, here.

Before starting the article, click here or below to view the associated sample map.

The linked map below, created by Robbie W. of South Carolina, was made with one click using DroneDeploy. Robbie’s drone took 222 separate images which were then automatically uploaded to the cloud and stitched together to form the final product—a single, sprawling 84.94 acre map:


As we learned in part one, photogrammetry is often more art than science

There are multiple factors that help to determine the quality of a drone map

As your drone flies its mapping course and takes photos, each image contains features like: crop rows, trees, buildings, trails left by equipment, or anything that is distinctly recognizable in the visual space.

As the photos are taken during the mission, the different features are captured multiple times, from various angles.

In the DroneDeploy example map above, the concentric crop circles in the field were ideal visual features for stitching

These features are identified, matched by a mathematical process, and aligned on top of each other to create one high-resolution map.

This is called stitching.


Examples of visual features that are ideal during the stitching process:

1 of 2: Shown are three identifiable visual features that help image processing software match images together during the stitching process
2 of 2: The same identifiable features as above but captured at a different angle in this second image

Stitching, however, is not simple.


Usually, humans are incredibly adept at pattern recognition

Give us a puzzle and we can figure out how to put it together by making comparisons and finding patterns. Even if it’s through trial and error, we will eventually solve whatever sick, twisted puzzle pretty much anyone could come up with… except for one.

Briefly consider putting together a puzzle of an aerial view of a city block…

In this puzzle:

  • all the pieces have undefined edges
  • all the pieces need to overlap by an unknown amount
  • there is a different perspective in each image that means the pieces have to be warped by an unknown amount to actually make them match

…now include the fact that vertical structures will look different from each aerial angle.

This puzzle would be nearly impossible for a human to get right

by Mike Warot, Flickr

Fortunately, leaving this intense photogrammetric stitching-process up to powerful computers is a luxury that we can all take advantage of and accessible aerial data can now be easily acquired by using various kinds of affordable drones.

But things are rarely perfect and all image processing software is not created equal. Imagine if you never even had to worry about any of this and your data was processed automatically in the cloud?


If all of your map data was stitched, stored, and processed in a secure cloud

You wouldn’t need to worry if conditions weren’t absolutely perfect when mapping.

As we’ve said before, photogrammetry is an art.

Something as insignificant as a single cloud has the ability to throw off the map stitching process entirely.

If every day were perfectly cloudless and sunny, we wouldn’t need to say much more but flight and mapping environments are always changing. We have our wonderful atmosphere to thank for that.

The Earth giveth and the Earth taketh away.

On those days when conditions aren’t ideal, it’s nice to have a flexible partner that you can trust to get things right.

Check out this improvement of the less-than-desirable map on the left:

The dark areas on the Original map were caused by clouds overhead as the images were taken from the drone. This caused errors in the stitching process. The Improved map was reprocessed, dark areas were removed, and overall map quality was improved.

The unusable map on the left was automatically transformed into the useable map on the right by DroneDeploy, without any user interaction. No extra app installs, downloading software updates, or experimenting with settings.

Because DroneDeploy is cloud-based, we have the unique ability to tweak settings and try alternative processing methods in order to give users the absolute best drone maps possible — without ever compromising the original data.

If intervention is necessary, DroneDeploy users’ maps can be transformed into the actionable pieces of data they were meant to be.


So yes, the stitching process is not simple but it is a relatively easy concept to grasp.

The adage ‘bad data in, bad data out’ is especially relevant in regards to stitching — but as seen above, can sometimes be mitigated if you choose the right drone mapping partner.

We all know that every industry’s data requirements differ, but doing everything you can to acquire the best images and working with the most flexible image processing software is an essential part of getting the only final product that all industries need:

Reliable data


Stay tuned for the third and final part of our series on ‘Making Successful Maps with Drones’, coming soon.

Happy mapping,

Click here to read part 3

Ready for more?

Check out our website at www.dronedeploy.com and sign up for free, here.

You can check our latest drone and device compatibility here.

DroneDeploy's Blog

DroneDeploy is the leading cloud software platform for commercial drones. Learn more at www.dronedeploy.com

DroneDeploy

Written by

DroneDeploy is the leading cloud software platform for commercial drones.

DroneDeploy's Blog

DroneDeploy is the leading cloud software platform for commercial drones. Learn more at www.dronedeploy.com