3D point cloud optimization

[adalta]

I am currently undertaking my Masters in Forestry. One of the main topics I am exploring is 3D surface fuels modeling in forested environments (open canopy---->semi-closed canopy). Given that the camera will be standardized across all test environments, my question here is what do you think are variables that will impact surface fuels modeling. When I say surface fuels I am talking about anything that is not tree canopy or tree stem. It has been shown in many research articles that one of the main driving forces behind wildfire spread horizontally and vertically is surface fuels. Some variables with flight planning that I can think of off the top of my head would be height AGL, cross-grid vs. single pass, time of day, shadowed vs non-shadows. The area I struggle with is what processing parameters will impact the results. I will say that the products that I am outputting now compared with 1 year ago "look" better, but how would you measure accuracy and what does "look better" mean? My idea was to add some sort of PVC box to the acquisitions that is a know size and volume and see how much better/worse each acquisition gets. The study area(s) I am thinking of using have had multiple LIDAR acquisitions so my thoughts there were to use that as a tool for comparison. Only downside is some of the acquisitions have already happened so using the PVC box as an accuracy measure would not work. What I do have is GCP targets as well as many tree metrics such as height, diameter and a few others, does it seem logical to be able to use natural features as accuracy measurements?

I have notice that more photos in photoscan does not mean better results, I was trying to model a 1 hectare plot and ended up with almost 300 photos taken from a 100m AGL, NADIR cross-grid and 400 photos taken from a 90m AGL, 10 degree oblique camera angle cross-grid. The reason for so many photos was we also had a micasense rededge on-board and the mission was planned for that camera resolution so the photos I used were for a much higher resolution camera where basically every photo was the entire plot. I figured what the heck lets just run all the photos, mind you I was not using a single machine and has a network cluster of about 8 very nice computers, but it failed just to many photos of the same scene. I have yet to re-run it with fewer photos. Do you think any sort of oblique photos will impact how well the surface fuels will be modeled under the tree canopy? I know google has played with this type of oblique photos for their new 3D google earth maps of certain areas, do you think it will help in modeling surface fuels?

[Alvaro L]

Oblique images are used for urban enviroments where you need to sample also lot of vertical façades. Besides, oblique images can increase your task budget quite a lot. For forest sampling I think nadir images will do, you just need to use a good sampling strategy with good overlapping but also one that really adapts to terrain (and forrest ?) heights instead of sampling from a single plane height, which is something you will be able to do only with drones and good software. Sampling from a single plane height a varying terrain or forrest mass is calling for all sort of insufficient overlapping problems. Also you can adapt routes to sample forrest boundaries, if you are not flying a plane but a drone you can think and fly out of the grid. Remember that part of your nadir focal is actually sampling in an oblique way.