I’m new to Pix4Dmapper and the biggest issue I’m having is reflective surfaces, like windows, and dark surfaces, like asphalt and shadowy areas, leave big holes in the point cloud.
Is there a way to change the settings to increase the generated output on surfaces like these.
I’m not sure where to begin and can’t seem to find any how-to’s on the subject.
Thanks
Hi @ctwbim, I know what you mean, it is challenging to reconstruct areas like that as the same area/feature is captured differently on each image due to reflections.
If possible, try to adjust the flight plan and increase the overlap and decrease the flight height. We did some tests and saw, with senseFly Aeria X camera, the reconstruction of asphalt was optimal when using the 75/75% overlap and GSD between 2 cm and 2.5 cm.
It also helps to fly on a cloudy day.
As for the processing options. If you would simply like to generate more points, then you consider increasing the Point Cloud Densification option Minimum number of matches to 2. But keep in mind that this can generate a lot of noise.
If you want to make sure that the holes are not visible in the mesh and DSM, then you can consider adding surfaces. They can be treated as constraints and will will the holes in the mesh and the DSM.
I hope this helps!
Thanks for your suggestions.
We had done most of these with a 90% overlap with the same results. The parking stripes would show up but not he asphalt between. Adding a surface isn’t viable since we’re trying to capture the topography in a point cloud for further processing in other softwares.
I was hoping there was a setting like on our terrestrial laser scanner known as “dark point filtering” where you can set the luminance threshold to exclude. I was hoping there was something similar in the Mapper software.
I guess we’ll stick to flying on cloudy days for now.
Thanks again!
Hi,
I’m having problems with the reconstruction of homogeneous colour walls. These walls leave holes in the point cloud and I don’t know how to fix it and fill these holes. I’ve already tried with the multiscale option but it doesn’t improve the point cloud.
Any suggestion?
Thanks
I wish I had some suggestions!
It seems to be the biggest point of contention for us with the Pix4D software. There’s just no happy medium between light and dark surfaces. If you set the Capture software to Sunny, you don’t get dark areas and if you set it to Cloudy, you don’t get light areas.
I’ve noticed varying results based on the quality of the drone camera.
Increasing the camera angle helps a bit for vertical surfaces but generally dark or reflective surfaces never show up well.
People have suggested creating a plane but that kind of defeats the purpose, in my opinion, when you’re trying to create an accurate point cloud.
We typically now try to fly missions when the sun is directly above or on really cloudy days as someone else suggested to reduce the number of shadows.
If you find something that works, please post back.
Hi CT and Francesco,
When dealing with large variations in light the best places to look are the exposure and the sensor.
With exposure, one needs to find the balance that allows the most detail in both the highlights and shadows. However, it is better to skew exposures towards more detail in the shadows than in the highlights. It is easier to recover overexposed highlights(detail is there) than try to pull detail where there is none. The range of exposure captured between highlight and shadow is also dependant on the quality of the sensor. Larger/better sensors can capture a wider range, while for smaller/cheaper chips the range is a lot tighter. As mentioned, capturing during times of diffused/cloudy lighting or midday with the shortest shadows can achieve the best results. However, I would caution against midday, as that is going to give you higher contrast between the flat/terrestrial surfaces, and the vertical facades that we are trying to improve within this question. The brief moments of diffused light in the morning and evening on clear days can provide moments of more even lighting as well. If the hardware allows, you could also ride the exposure, changing with the light and or subject matter. It takes a little practice and is best with multiple operators, but can get the best results in tough lighting situations. You could also use one exposure for nadir flight paths, and another for oblique, each optimized for the surface or facade.
I’ve noticed varying results based on the quality of the drone camera.
Higher quality sensors are going to give the best results as they have a higher dynamic range. It still can be hard to find the right balance in extreme lighting situations but they still give more room for success. With a wider range, they will capture more information giving the software more to work with for better reconstructions.
Have you considered merging projects with aerial and terrestrial capture? Using PIX4Dcatch or still camera datasets merged with aerial imagery can give you more detail in the facades as well as from dark areas. PIX4Dcatch can also incorporate LiDAR data from the latest iPhones/iPads enabling better reconstructions from traditionally difficult subject matter where photogrammetry alone has struggled. i.e. large blank surfaces. The article Depth and dense fusion can help illustrate this advantage.
One last point to touch on would be a proper flight/capture plan. The article Step 1. Before Starting a Project > 1. Designing the Image Acquisition Plan offers optimal flight plans/best practices according to the subject matter, and can provide useful information.
I hope this information was helpful. I am also curious if there are any other suggestions out there as well.
-Jon
@Jonathan_Dennis Your post was most helpful and provided insight into additional strategies to incorporate. Much appreciated.
May I ask you or someone at the office to influence the Pix4D web developers to undo the travesty of hyperlinks now being useless in these posts? Every hyperlink (except marketing) is redirected to the generic Pix4D Documents page without referencing the specific article or page previously cited.
Every link redirects to the “Uh-Oh. This page knows nothing.” Pix4D Documents page, which is utterly worthless for those attempting to find solutions to their issues.
I have attempted to use the search feature on the Documents page, but the output is neither user-friendly nor useful, and I feel that these community gems are being lost forever.
Thank you.
Michael.
