(Ps. Due to attachment size limitations additional attachments referred to in the below text were added in seperate replies below the following post).
Hello all,
I would like to get some advice regarding the following. Earlier this week we performed a test in order to produce a 3D model of a small (100m) former traintrack bridge (see attached image: Image12_Bridg_Plan (left)). Having no prior experience we expected somewhat of a challenge, and a challenge it proved to be indeed! We initially executed an automated nadir grid flight, followed by a cross-grid flight over the structure with the camera pointed down at 45 degrees. We found that lightning conditions varied a lot (due to vast amounts of sunlight and changing terrain underneath the bridge (water -vs- vegetation -vs- soil). Hence we decided to let go of the automated flights and perform a series of manual flights and also to trigger the camera manually. This resulted in the following image sets:
1. Nadir imagery: Approx 8m ABL (Above Bridge Level)
2. Horizontal imagery: Approx 5m from either side of the bridge (camera at 90 degrees)
3. Oblique imagery: Approx 5m from the bridge, 4m ABL (camera at approx 45 degrees)
4. Oblique imagery: Approx 4m from the bridge, 8m ABL (camera at approx 30 degrees)
5. Oblique pillar imagery: Approx 4m orbit around pillars (camera at approx 20 degrees)
All of the above values are estimates as the flights and camera orientation was done manually, but see the second attached image (image12_Bridge_Plan(Right)) for a more comprehensive overview.
As an initial test we tried to align all of the images at once, using High Alignment settings, and both Generic and Reference preselection turned on. We did not apply any masking at all. The output was useless, resulting in a highly dispersed sparse point cloud. It seemed that each flight line produced a sparse point cloud of the bridge structure of its own, but none were aligned with one another (see attached: image3_misalignment_total). We ran similar Alignments using only Reference preselection, only Generic preselection and no preselection; but differences were slim and the model remained useless.
Regardless of performing a manual flight and adjusting lightning conditions (EV value) for each image, illumination differences sustained. Consequently we then performed some editing in an attempt to produce more homogeneous images. Unfortunately with no succes; the model failed to be build properly. Even when taking out the orbital flights around the pillars out of the equation, the results were the same.
In an attempt to diagnose where things went wrong we decided to process each of the flight lines/ camera angles seperately, for each side of the bridge. Unfortunately still then, apart from the nadir flight and the orbital flights around the pillars, none of the image sets aligned properly. Even a set of 6 horizontal images of one side of the bridge, which was very properly illuminated in this particular instance, failed miserably. Even though it is stated all images were aligned, the order is off by A LOT. The order of images after alignment is in no way correct with respect to the order in which they were taken. The last image at the fartest away location, was actually projected on the location of the second image, and vice versa (see attached: image4_misalignment_horizontal). It seems then that Photoscan was not only unable to align images in between different flight, but even alignment of images within a single flight line proved troublesome and useless.
This was also indicated by relatively few projections per image after initial alignment, varying between 84 and 540 (even though Alignment was set to High and Key- and Tiepoint limits to 160k/40k). In a final attempt to aid in the alignment and solve the issue I added at least 4 manual markers/ tie points on each image, as well as on subsequent images that they were visible on (see attached: image5_manualmarkers). However, after either resetting and re-aligning a subset of images OR after re-aligning all images, the misalignment remained completely. I am wellwilling to share the (only 6) horizontally taken images of this small subsection of the bridge to allow others to give it a shot!
So, what’s next? As this is a highly important test for us we are forced to redo it during next week. We will visit the location once more, during overcast this time, and collect massive numbers of images of a small (20m) portion of the bridge from even more positions and camera angles. Hopefully resulting in a sufficient number of flight lines and images that will produce a succesful model. Afterwards we can iteratively take out flight lines/ camera angles to arrive at an optimum and minimum number of images and camera angles required.
Any help and advice on taking on this challenge is much appreciated! What should we have to look (out) for? What are the do’s and don’ts? How many flight lines should be incorporated, from what distance and position with respect to the bridge? What about camera angles and minimum overlap? Regarding oblique zipline flights along either side of the bridge, should be change the lateral distance to the bridge as we make the camera point downward more and more, or should be keep only vary altitude but keep the ziplines aligned in X and Y? Could it be that we flew too close tot he bridge and should we perhaps acquire imagery from a larger distance? And what about mission planning? Provided that we can indeed fly during overcast, should automated flight work or is manual flying and illumination adjusting still necessary?
Topic: Advice seeked: 3D Modelling of a Bridge (Read 6071 times)
