Questionable DSM output: Noise, artifacts and weird shaping

Has anyone else been experiencing troubles regarding how their outputted DSM appears? Particularly at the edges of objects as buildings.

I have been running several test runs in the past weeks using a nadir dataset taken from 50m AGL, resulting in a GSD of approximately 1.25cm. For reference purposes only I enclosed a screendump of an orthofoto created on the particularly troubled region:

I had first be running some test with the previous version of the software. Settings were largely set to the default specifications for 3D models. The below DSM was outputted using IDW and Sharp for smoothing;

Although from a distance the DSM looks rather okay, upon closer inspection I find that the edges are way too curvey, almost cartoonish. The edges of objects, including rooftops and benches, are far from clear crisp. I was then advised to rerun the project, but this time using Triangulation instead of IDW. This time, the output was rather noise and thus I went on to remove artifacts around building’s edges as suggested by the Knowledge Base Here’s what I got afterwards:

The curvy appearance of edges, as shown for IDW processing, have seemingly dissappeared. This, however, was replaced by substantial amounts of noise. This noise is particularly visible for the most northern building for which no artifact removal was done. But even for the buildings in the center, for which artifacts were supposedly removed, noise still continues to exist. 

Upon contacting Support again I was told I should give the latest version of the software a go combined with Triangulation, as the DSM generation supposedly was improved a lot. Below is the output I generated thereafter, the first using Sharp smoothing and the second using Smooth smoothing:

Again, the output is highly questionable. More importantly, it is fully unusable for measurement purposes as one can impossibly define the true geometry of buildings due to substantial noise. I have run similiar trials using a different software package, and the same dataset, and strangely enough found that the DSM seemed to approximate the field situation more realistically.

Has anyone else been experiencing any of the above, let alone be able to provide some suggestions on how to circumvent these issues?

Dear Bob,

We got similar results to yours. When looking at the project, we see a correlation between these artefacts and holes in the point cloud. The sides of the buildings are also not very well reconstructed. This is due to the images being taken nadir, which means that the sides of the buildings are not visible on many images, and most of the time at sharp angles. 

To get a better DSM for this type of area, we would recommend flying a double grid flight, with the camera at an angle. This should give better overlap, and better imagery of the sides  of the buildings. The vertical walls being better reconstructed would translate into a sharper DSM. 

 

Best regards,