Public Lab Research note

3D model in MeshLab

by cfastie | May 31, 2012 14:19 31 May 14:19 | #2275 | #2275

I tried a few more times without success to display the Hypr3D model from the Lee, NH cover crop trial balloon mapping photos in MeshLab. Above is a snapshot of the MeshLab display of the low resolution model from Hypr3D. The high resolution version will display the model but without the photo texture mapped on it. Below is the model in Hypr3D. The icon at the upper right will display it full screen.

This is a new version with fewer photos (155 out of 159) and some brightness matching.

You can download the high or low resolution model here and try it yourself.

Does anyone know how to make the HighResTexture.jpg display on this model in MeshLab?


This is pretty amazing Chris, how do you see it being useful for farmers? It seems to me that this could be useful to people interested in mapping Hydrogen Sulfide or other gas concentrations as the addition of typology could give a clue as to why you have differing concentrations of Hydrogen Sulfide in a given area. Also it might be helpful to develop a step by step how to guide for this? Exciting work!

Is this a question? Click here to post it to the Questions page.

Reply to this comment...

Hi Sara, I think the original idea with this model was that it could be used to calculate the volume of vegetation (above some ground plane) of each of the 52 study plots. This could then be related to the biomass of the plants there which could help reveal which treatments had an effect. I am not sure that will be very easy to do because the model does not have horizontal dimensional control -- note that the rectangular plots are longer at one end of the study area than the other (they should all be the same size). It might instead be possible to use many random measurements of height of the plant canopy in each plot and relate that to plant growth. Deriving such quantitative results requires calibration, so maybe poles of known height could have been placed throughout the field before it was photographed (although that might not really work).

Regardless of whether quantitative results come from this, it is a great way to display the geometry of small landscapes. Topography, geographic features, vegetative structure, and the built environment can all be displayed in "3D" either on the web or during a presentation. Most people, even those who do not relate to maps, get the picture very fast when they can fly around a landscape. So an important value of this tool is as an effective way to direct an audience's attention to some feature of a 3D space. It can also reveal details of geography that might be overlooked on the ground or from much higher altitudes and could have application in lots of different endeavors.

Reply to this comment...

I was thinking that we could try to produce a false color image to show the topography (basically a DEM or digital elevation model) and then adjust the false color to be relative to the average of the four corners of each respective plot (to account for overall topography change). Then we could rectify the whole DEM to a measured grid using Photoshop or something, and calculate volume by looking at (the height of each pixel * the real-world pixel area) and summing them up for each plot.

It's several steps but it seems like a viable way to approach biomass calculation.

Has Dorn done independent ground-based biomass measurements to compare to? I'll email him to see if he's following this thread.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...

I think Dorn was planning on measuring biomass sometime during the next week. He will harvest all plants in two 1m2 plots in each of the 52 plots. You could do the same thing digitally, with more sample plots, so there is no reason to rectify the whole image to remove the random horizontal distortion. However, the same variables responsible for the horizontal distortion could also distort the surface vertically. Without calibration it would be hard to defend your result.

Here is a snapshot of the highest resolution model I have produced so far in MeshLab. This is from a portion of the model that used the lower elevation photos (some parts of the model did not). It does not have the jpeg texture mapped on it (because I don't know how), but that is irrelevant to volume calculations.

MeshLab model of cover crop trials

There is lots of noise in this model (the troughs are four foot wide mowed paths that were pretty flat and level), so determining a base plane from which to measure volume in the plots between the paths will be tricky.

Reply to this comment...

Thank you Jeff for pointing me to this thread. I think Chris represented the on the ground data collection well. I am really excited to have the opportunity to gather this new richer data set I never anticipated having access to, and joining this community! I am thrilled that this Labs collaboration happened so soon and so smoothly.

We will be cutting, sorting and drying samples from each plot in the next week or so. With that data we could potentially do some color and volume calibrations. One nice thing is that we do have known plot sizes which are clearly visible (8x16) and the bottom of the mowed pathways should identify the ground level - but as Cris said, even in the higher res version there is a bit of noise.

However, it seems that a percent cover calculation of vetch would be potential low hanging fruit because the data would be relative rather than absolute and the color differentiation should be fairly clear and does not depend on the 3D rendering.

One of the great limiting factors for doing more on-farm trials is the labor involved in data collection. If we can figure out a method for remote data collection and basic automated processing, it would open up a lot of potential for consistent localized, applied and producer driven research. I have shared this concept with several agricultural professors at UNH and they are very excited to follow up on the potential for invasive plant species monitoring and documenting weed pressure in plots. The folks at Cornell are interested in the NDVI for their adaptive nitrogen field trials.

I look forward to the conversation.

Reply to this comment...

Hey Chris,

Great work! On lower-memory machines MeshLab often struggles with large texture files like the ones we create on Hypr3D. If you experience this problem you can work around it by resizing the texture file to either 4096x4096 or 2048x2048. This will of course reduce the texture resolution, but it's often a minor effect. Thanks for using Hypr3D, and keep up the interesting work!

-Tom CTO & Co-Founder

Reply to this comment...

Tom Milnes from Hypr3D responded on the discussion forum with a workaround to the problem of the HighResTexture.jpg not displaying. He recommended downsampling the jpeg because the high resolution version requires a lot of memory to map and display it. I downsampled by 50% and it worked. Thanks Tom. Below is a snapshot from MeshLab of the high res model with the 50% res texture map.

Model with low res texture

If anyone with more than 8 GB RAM and 4 GB graphics memory wants to try the high res version, you can download it here and open it in the free MeshLab. Nathan Craig made a video about how to do this (with an older version of MeshLab).

Is this a question? Click here to post it to the Questions page.

Reply to this comment...

All the biomass samples were collected this week and are being sorted, dried and weighed at the lab. I am taking soil samples today or tomorrow. Within a few weeks to a month we will have all of the data entered electronically - which I will post.

Reply to this comment...

Login to comment.