News & Updates

Photomesh 7.3 makes big improvements to true-ortho quality

We take an interactive look at the improvements to aerial photography processing using Skyline Photomesh.

Following on from our previous blog, we’re looking at the improvements in true-orthophoto technology.

The most recent Skyline Photomesh 7.3 update makes some big gains in the way it produces high quality orthophoto products. This is particularly important for our XCAM users using Track and Circle pattern surveys.

Now we’re not saying that the orthophotos we have delivered in the past were bad, far from it, but this aims to show that the world of photogrammetry is rapidly evolving and improvements are always there to be made.

To produce a true-ortho the processing software takes the textured 3D model and flattens it. Using this technique in the past has thrown-up a number of graphical issues that make the imagery harder to interpret as it’s not a single snapshot, but a mosaic of many images stitched together. In this analysis we’ll see whether these issues have been resolved.

 

Methodology

For this brief study we’re using an XCAM survey captured in a circle pattern, a 5cm GSD dataset from Bath, UK from March 2017. The original orthophoto was produced in Skyline Photomesh 7.0. We’re using this as our ‘Before’ dataset. 

We’ve then taken the exact same project and reprocessed it in Photomesh 7.3 to produce a new orthophoto. This will be our ‘After’ dataset.

 

Moving Objects

One annoyance with the old version of processing was that it would often mosaic images together in inconvenient places. You would regularly find cars and buses chopped in half where two images taken at different times had been stitched together. This happens when a car moves between two images being taken. The software tried to use the best image for each part of the map but didn’t take into account the time between images.

Although this is inconsequential to the usefulness of the data, it is distracting and can be confusing to the casual viewer.

Here are some examples. We can see that in most cases the vehicles are now complete objects. One by-product of this is that there’s a higher chance of the software including a vehicle than trying to exclude it. So there’s a trade-off but the overall benefit is a more true-to-life scene.

Use the image slider to swipe between the Before dataset (Left) and the After dataset (Right).

 

Building Halos

The software is producing the true-orthophoto based on the 3D model it produces in a prior step. This means the sides of the buildings aren’t visible (as discussed in a previous blog). However, because the 3D model is never completely perfect, the edges of the buildings aren’t as well defined and straight as they could be.

The result is a halo effect, a protrusion from one of the faces or a smear in and around the building. Basically ‘wobbly building syndrome’. It’s not nice to look at and if a user was deriving information (such as a roofline) from the data then mistakes could be made.

With the new software we see far less building halos and wobbly effects on the buildings. They have much more straight-edge lines and well-defined ground features, even on this older dataset.

Again, swipe these images to see the difference between the Before (Left) and After (Right).

 

Roof Detail

Another aspect of building detail is in how Photomesh renders chimneys, skylights and dormers. As these are quite intricate detailing on a building they’ve not always been the easiest for the software to produce. It’s useful to know whether a building has a chimney stack or a roof extension so seeing that clearly from an orthophoto is important.

The improvements here are being able to see individual chimney pots and crisp rooflight detail on the image. There’s certainly been a big improvement here. Much less of the scene is blurred.

One caveat on these examples is that this dataset isn’t the highest resolution XCAM can produce, so we would expect better results with other, newer datasets.

 

Vegetation

Our users have very simple requirements when it comes to vegetation; trees should look like trees. The previous effort relied too much on the 3D model to render the trees and if there were imperfections in the model they would translate to the ortho.

The new 7.3 version of Photomesh does a much better job. The canopy of the tree is true-to-life in most examples and you get much less ‘jelly-mould’ shapes. One by-product is that the old Photomesh did ‘see through’ the trees to the ground, displaying more ground information such as yellow road markings. But we think this is a good price to pay to get a better looking ortho. And if underneath the tree is important, then the user can always use the 3D model instead.

 

Water

Water is photogrammetry’s arch enemy and is always a challenge when producing an ortho. Previous versions would show a range of colours and shapes representing what should be a flat water surface. The new Photomesh has much fewer visible cut-lines and improves the overall aesthetic of the aerial photo.

 

Object Removal

One final thing that has been a great improvement; rendering of small ground features. For our customers the aerial photograph is a piece of evidence to use in their GIS tasks so they would expect to see something on the ground if it was indeed there. In a few rare cases we have seen objects removed by the processing software, which isn’t great for showing a true-to-life scene. The software now does a much better job of rendering these thin objects.

This example below shows the improvements made here.

 

Summary

Overall, we’re now able to produce a higher quality aerial photography product that can be used to extract information and derive understanding of the landscape. We are in the process of re-delivering a few datasets (including this Bath dataset) to customers so they get added benefit for their existing orthophoto products.

GeoXphere and the XCAM product is software agnostic when it comes to aerial imagery processing, but we firmly believe in using the right tool for the job. We have seen the huge improvements Skyline have made to the Photomesh product over the last year or so and that’s why we use their software for our own processing and demos. We can also bundle Photomesh into an XCAM lease so XCAM users have all the tools they need to produce high quality 2D and 3D products.

 

For more information or a free trial, please contact the GeoXphere team and we’ll be happy to help you started with Photomesh, XCAM or XMAP.

 


  • +44 (0) 2034 114 544
  • This email address is being protected from spambots. You need JavaScript enabled to view it.
© 2017 GeoXphere LLP. All Rights Reserved.