Hello,
We are trying to improve our workflow over dense vegetation areas. I found one article that gives general cues but I want to further my comprehension of other parameters because I have read contradicting things.
Step 1 > General > Keypoints image scale:
From what I understand, the lower the scale, the less number of keypoints is generated (because the algorithm limits matching point search to images that share close location attributes) but the more confident you can be about those keypoints. However, less points means less the precision of the model. Why would a lower scale be recommended then ? I would have though you always want as much keypoints as possible, so why using a lower scale improves project when you have a lot of uncalibrated images?
Step 1 > Calibration > Method:
I have seen many posts where alternative method is suggested over Standard calibration. However, as I understood it, the alternative calibration method implies that you have good trust in your orientation metadata… but isn’t that what accurate geolocation is for (aka you need RTK drone)? So how does Alternative differ from Standard?
Step 2 > Point Cloud > Image Scale: same questions than for Step 1 / images scale
Step 2 > Point Cloud > Multiscale: I have seen post where they do not recommend activating that parameter when working in forested areas. Or that the option introduced more noise. Again, I would have though more points = more precise point cloud. Any thoughts?
Step 2 > Point Cloud > Minimum number of matches : Our flights always have very high overlap (minimum of 80% longitudinal and 70% lateral) but we are mostly in dense forested areas. If we use a minimum of 5 or 6, can we thrust to have more accurate point cloud (aka more accurate DSM?) ?
Step 3 > DSM and Ortho > Raster DSM Method : the same article sited earlier suggests to use Triangulation method for improved dsm in forested areas. However, Pix4D says on its website
So why would I use that?
I thank all this community for their welcomed feedbacks!