Support Website Contact Support Blog

Large Differences in GSD?

I’ve read that for a large site where the topography has large changes in elevation, if you fly the same altitude AGL from the starting point, it can create issues in processing.

However, if you have a target GSD, and therefore a target altitude AGL, as long as you never exceed that altitude so that the variation in GSD will always be better GSD than the target, it would seem to me that shouldn’t cause any issue.  The issue would be if the GSD was a lot worse than the target at certain points in the flight path.  Or is the potential issue in processing just any big difference in GSD despite whether it better or worse resolution?

Example:  I want to generate a topographic map (DEM/DSM) where I want the elevation measurements to be accurate within 12 inches.  It is widely stated the elevation measurement are *typically* accurate within 3 times the GSD.  So I need a GSD of 4inches/pixel.  (In reality it will be better because that GSD allows for an altitude above 400ft).  So if some of the photos have a GSD of 1inch/pixel because of high points on the site being surveyed, whereas others will be 4inch/pixel at the low points of the site, is that going to create a problem in processing?

 

Hey Jeff,

I believe after you are done processing your GSD would stay exactly the same across the whole field. Alternatively you can force Pix4D to down-sample your GSD to a lower resolution if you want to make sure that your GSD is consistent everywhere. 

I think you may have misunderstood my question.  The work products I referred to are three dimensional. GSD is just horizontal resolution.  I’m wondering if processing photos (from the same camera shot on the same day) that have greatly varying GSDs due to changes in the ground elevation while flying at a constant altitude would negatively affect the accuracy of the geolocation coordinates produced after processing.

 https://support.pix4d.com/hc/en-us/articles/202558889-Accuracy-of-Pix4Dmapper-Outputs#gsc.tab=0

Generally, one can expect an error of 1-3 times the Ground Sampling Distance (GSD) of the original images for the relative position of a point in a project that is correctly scaled and reconstructed.

 

Hey Jeff,

Opps yeah I didn’t realize that you were referring to a three dimensional  product. I though that you were trying to get a 2D orthomosaic/DSM and asking about the GSD of that. I have never processed a 3D output so I am not sure exactly but if you want to increase the accuracy of the geolocation you may try use GCP’s if possible or MTP’s.  

Hi Jeff and Selim, 

A large difference in Ground Sampling Distance (GSD) can create problems in the reconstruction, which is why it is recommended that the largest GSD is not more than twice as big as the lowest GSD in the project.

To understand this answer one must look at the basic workings of photogrammetry which happens during step 1. Initial Processing in Pix4D software. In each image keypoints will be identified at the GSD of this pixel. Once keypoints are found separately in each image, the software will try to match keypoints between images. That is where a large difference in GSD can be a problem, because if one pixel represents more than twice the area on the ground than the other pixel, it will be hard to match them. This means that the question of having accurate coordinates with a pixel can sometimes not be asked, because there might be problems in the reconstruction before that. This is why it is recommended to have a homogeneous GSD over the project area.

In addition, the coordinates associated to a pixel cannot be more accurate than the GSD of that pixel, e.g. if a pixel represents 10cm on the ground, then you can have measured and associated coordinates with an accuracy of 2cm to this pixel, but it is still not going to be more accurate than the GSD of the pixel. Let me know if that is not clear. 

Here I would like to mention that the GSD that is reported in the quality report is the average GSD over the project area. This means that some areas can be more accurate than others. 

These two links provide more information on the subject: