I am requesting expert advice in determining camera field of view and data processing considerations, for the purposes of proximal Structure from Motion 3D volumetric reconstructions of 0-2 m tall plants in an agricultural field, taken from a 1.5 mph moving platform, using color cameras and the pinhole optical model compute?
This because in Maricopa Phenotyping, Plant Group Phenotyping Team 2019 experimentation under the leadership of Dr. Thompson, we plan to image cotton plants, using three Nikon N1 aw1 16 MP DSLR action cameras, triggered together at 1 Hz, one camera mounted in nadir view and two oblique on either side of the row crop. We plan to process images using Photoscan.
However, as final (adjustable) camera mounting positions are created via new square tube arms on Professor PSC, I seek additional input in how to suspend, point and set cameras optimally, and so support success in the expected subsequent large volume SfM processing.
Please see iteration two in the Project update, “A second year Professor – Tenure Track?”
https://www.researchgate.net/project/Maricopa-Phenotyping/update/5c0d42d93843b006754a3b1c
Do you want to move that position as time goes on to accommodate changing plant size, or one position for the season? Changing positions allows more accuracy at each size, especially if you have some known objects in the mix to standardize from... In our case, we are looking for pathogens, etc, so moving the cameras makes sense as we strive to image the areas we need to evaluate... this then requires a one size fits one system, which is a hassle to set up compared to a one size fits all...
I really appreciate this comment, Steve points out perhaps a most important consideration.
- Holding a constant sensor height to soil throughout the growing season allows a constant sensor FOV space, where the plant growth can fill. However distance from detector to target will decrease over time. – one size fits all
- Moving sensors upward in tandem with plant growth maintains a relative constant distance from sensor to target, while the sensor’s absolute FOV space increases over the season with higher mounting positions. – one size fits one
When oblique angles come into play, the distance from target change effect is compounded by additional angular ray offset, such as in the perspective optical model.
Additionally, proximal imagery that is collected within a few meters of space appears to acquire photographic and geometric considerations of significance different from a landscape level vision.
To answer the question, we will rise, because as offered, more accuracy is expected. Professor PSC is designed to have the sensor system adjust relative to plant canopy average height and therefore maintain a constant detector to target distance zone. We do this on the Avenger rig as well and with the same spatial theory with phenotyping tri-metric data, note that this piece is especially important when developing vegetation index over seasonal time.
- Spatial physical visual anchor points involving objects of known size and reflectivity allow image rectification and biological component measurement.
Thank you expert Steve West RD4AG, I honor your experience and hard work serving the agricultural and research community, which goes a long way, rippling outward sir.
How about positioning a phenotyping camera under the plant canopy and pointing it up, into solar illuminated leaves?Presentation In-field leaf transmittance imaging for phenotyping
So far I have not been able to direct the SfM compute with discrete FOV projections as EXIF defined by populated GPS and AHRS coordinates. I would like to first spatial match all tie pixels then filter and process the point model using custom functions. If you can compute 17K proximal field plant pictures as a single collection event data object please contact me.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images