Hello all, I've been working on a project at my university that employs a surface water robot as the means for collecting data on a polluted New York City canal. I started work on a camera system for the robot that would allow us to create VR images for the project, and hopefully google views as well.
The result is a motorized rig that positions a single DSLR (Nikon D7000+ Samyang 8mm f/3.5) to collect 6 photos, a nadir, 4 shots around, and a zenith. The rig is constructed of two milled aluminum plates, with various physical adjustments to align the NPP, two dc planetary gear motors, a worm gear train for the tilt axis, a belt pulley for the pan axis, and a combination of encoders and an arduino micro controller to program the rig, along with a simple GPS cable release to trigger the shots.
Here's a little video to give you an illustration of what I'm talking about:
https://www.youtube.com/watch?v=94JSpMMWh0g
And here's one of the resulting panoramas from our most recent testing:
https://www.360cities.net/image/gowanus-…0.00,0.00,100.0
I'd appreciate any advice you may have for improving the quality of these panoramas. There are several adjustments I'll be making for the next outing this weekend, including changes to the camera and rig program.
One of the first things you can see in both panoramas is the clarity of the top/bottom areas where the photos are taken while the camera is stationary, contrasted with the lack of clarity around the mid section where the 4 pictures are taken with the camera in motion. My goal is to both improve the quality of the stitches, by minimizing the time it takes to complete one cycle and reducing the relative motion in between the 6 photos, as well as keeping the image quality as high as possible, by using low ISO, an aperture smaller than f/5.6, and of course a shutter speed high enough to freeze any motion. I used a shutter speed of either 1/250 or 1/320, which I believed to be enough to negate the effects of motion while capturing the images, but in looking at the objects in the distance, fences, buildings, etc, they really are inexcusably fuzzy. This could be due to the focus on the camera, which I believe had been set around 1.3m, trusting in the depth of field of the fish eye to keep those objects in the distance sharp enough. In the underbridge panorama , you can see the clarity at the zenith degrade suddenly when looking at the edge of the bridge where the moving images are sourced. I'm trying to determine if the focus or motion blur is to blame for the lack of clarity on the distance, so I'll be try adjusting both of those next outing.
I am currently using PTgui for all my stitching needs, and will be upgrading very soon so I can make use of the masking feature in order to remove the aluminum side of the rig and force the nadir to cover all of the robot visible (I have to source one of the 4 to cover the area blocked by the aluminum plate during the nadir). One issue I am having intermittently is two small blank triangles appearing where PTgui is unable to find any image overlap to cover the panorama. This doesn't seem to be happening often, and id be surprised if there is enough variation in motion of the rig pano-to-pano in order to be missing overlap sometimes but not always. I know that the offset of the zenith to row shots is important to keep this from happening, but I believe this is set right. I'll attach sources/examples for this.
I'm still playing with the workflow of the rig, and will try testing a version of the program that makes a pause during each capture to eliminate any motion blur and allow for lower shutter speeds and ISO's, therefore better image quality in low light scenarios (under bridges, night time, cloudy days). If you have any suggestions in how to best program the rig, I would love to hear that as well.
Also, should any of you creative folk have clever applications in mind for the automated rig, lemme hear them!
I'm linking the source files for the underbridge pano, attaching an example of a pano with missing spots and linking the sources for that pano.
Source images of all sizes can be downloaded via flickr album here
Many thanks to anyone able to help be trouble shoot and thanks to all for building a great community that has helped me much already
