More photogrammetry from video data

Creating photogrammetry assets from high-resolution drone footage continues to be viable for quick results that don’t require high levels of accuracy. 

Two videos shot from a DJI Mavic 2 Pro over the suburb of Southern River in Perth, Western Australia, were used to create this model. 


The first capture was shot straight down, capturing a number of vertical aerial. While the second was shot in a diagonal orbit around the centre of the street. 



Using the method described in this earlier article, one image for every 15 frames was extracted from the videos, creating two batches ready for alignment in Agisoft Metashape.



Batch 1 (top down) contained 275 cameras and 29,216 points while batch 2 (orbit) contained 122 cameras but 36,925 points. The merged chunk produced the results seen in the video above and the images below.




The model is viewable on Sketchfab, though it has been decimated by 75%