Footprints…
This year for St. Patrick’s Day I spent the weekend in Inishowen, Donegal, where the weather was unpleasant to say the least. For about 2 hours on Saturday, it was dry but seemed to always be a minute away from raining, so I headed for the beach without any of the proper equipment for outdoors photogrammetry. I did have my phone though, which can be hit and miss for 3D scanning as there are few manual controls on the Note 4, so if the sky is present in any part of a photograph, exposure goes crazy, making shots unusable. Also, it only shoots in jpeg so there is very little wiggle room on the texture side.
Three areas in particular caught my eye for some hasty shots as it began to rain- a sandy boardwalk descending from the car park to the beach, a rock about 3 feet in diameter that was exposed due to the tide, and some footprints perfectly preserved in the wet sand, which can be seen below. The beach was just damp enough to hold the shape of the impressions and the roughness of the displaced sand around each of them. All of the subjects were ground based too, which was a blessing while shooting with the phone as it kept exposure consistent between photos. I took about 45 photos in an orbit around the footprints, being careful not to compress the sand and disturb the scene. In the short space of time it took to take these shots it began to rain heavily, so I headed back to the house to process the images. I used the unprocessed jpegs straight off the camera, they seemed pretty clean and the cloudy conditions meant that there was no strong light or shadows to deal with. This time around I used Reality Capture for the photogrammetry, and it performed well, aligning every photo that I had taken which provided me with an 8 million poly high res model! Some decimation, texture cleanup and an export into Toolbag 3 yielded the results below. I picked an evening sky pano and stuck in a very low direct light to add rake shadows across the sandy surface, and a little fog. Inverting the albedo map gave just enough roughness to sell some dampness on the surface.
Under various PBR lighting conditions
If you’d like to check it out in real-time in your browser, check out the Marmoset Viewer scene here:
https://www.artstation.com/artwork/EX0xq
Very impressive work Pete! I’d like to get into photogrammetry myself (am planning on building an automated weeding system, would be ideal for that methinks).
No better man for it Dunc! You thinking of building a rig? I’ve seen some good stuff done with Raspberry Pi cameras
Thanks Pete 🙂 Indeed, I am! It’ll be solar powered with either a CNC type axis or robotic arms. Am thinking either laser or a mechanical gripper for the actual weeding. And RTK for precision. Multiple cameras (at least 4 – and yes, using a Raspberry Pi or two 😉 ) looking inwards and down on the bed being processed. Here’s a shot of some of what I have so far: https://twitter.com/duncanaitken/status/839766593546551297
I might also use AWS with OpenCV for some of the processing… that, or get an NVidia dev-kit for autonomous vehicles (but looking in, rather than out): http://www.engadget.com/2016/01/04/nvidia-drive-px2/
You’re already way ahead, keep me in the loop on your progress
Great ppost thank you