4
« on: July 23, 2019, 02:47:13 AM »
I'm testing Metashape to see if it can give the same quality of products as Pix4D, which will then be used for treetop detection.
Currently, we get great results from Metashape in terms of camera alignment and orthomosaics, but the issue is that the point cloud doesn't seem to have the same detail as Pix4D. The point clouds/DSMs from Metashape appear smoothed, as though blanket has been draped overtop the forest.
The standard workflow developed is for photo-alignment to be run the default settings (Quality: HIGH, pair preselection: REFERENCE, key point limit: 40,000, tie point limit: 4,000) and for Dense Point Cloud to be quality: Medium, and depth filtering: Mild. Those settings result in a lack of definition in these point clouds/DSMs reduces the accuracy of a treetop detection algorithm.
To try and improve the point cloud/DSM we also ran tests with more intensive processing (Quality: High, depth filtering: Aggressive) which took more than twice as long and still had the smoothed treetops. The smoothed DSMs result in extra hours of manual edits compared to Pix4D DSMs.
Wondering if anyone has any suggestions of settings that could improve the point clouds/DSMs to appear more similar to the Pix4D products. Thanks.
NOTE: The attached photo is of 3 snapshots of a DSM/hillshade of forest canopy. First is made with Pix4D, second is Metashape (Quality: Medium, depth filtering: mild), third is Metashape (Quality: High, depth filtering: Aggressive)