Camera Position for oblique aerial stereo pairs

[PAOBC]

I need clarification of the Camera Position parameters from someone out there in the community who works w/ aerial reconn systems. The parameters lat/lon/altitude/yaw/pitch/roll seem a bit generic but understandable given the variety of PhotoScan users. However in my case I need these values to be definitive given the capture orientation of my camera system. I'm attempting to generate a DEM from a true stereo pair. The pair are captured on the same leg of a mission, where the 1st image (FWD) is collected looking forward of the platform, the 2nd (stereo mate) is collected AFT of the platform. The temporal difference between the FWD & AFT collects is a function of the platform travel. The lens assembly is in a 2-axis gimbal where the pitch & roll of the lens (line of sight) are independent but move in tandem as each image is collected. My particular stereo pair collect is ~15 deg left of nadir in the x-track direction x either +6.5 deg forward for the FWD image or -6.5 deg aft for the AFT image. The stereo pair overlaps by ~90%. My issue w/ the camera position is how to properly define the parameters. For example are the lat/lon/altitude parameters for the platform or the images given the collection geometry of the images. Likewise w/ the yaw/pitch/roll. For instance should the pitch parameter be only of the platform or a combo of that + the lens pitch angle? I've tried all kinds of variations of the above, none of which resulted in believable angular or pixel errors after performing "align photos". I have common gcps (markers) placed in both images as well. I've read in the forum that PS has issues w/ long focal lengths. My camera has a LONG focal length. Input would be much appreciated. Thanks

[Alexey Pasumansky]

Hello Paul,

Usually it is necessary to have at least three reference points for georeferencing, it includes camera locations and marker (GCPs) locations, providing that cameras are aligned and each marker has at least two projections on the images from aligned image subset.

The coordinates of the camera locations should represent the camera center at the moment the image were taken. As for the yaw, pitch, roll angles - currently PhotoScan assumes that camera is fixed under the aircraft in such way: when the aircraft is flying to the North and East is by the right wing (Yaw, pitch and roll are all zeros), the camera is pointing vertically down and the image axis bottom -> top direction is pointing to the North, whereas left -> right vector is pointing to the right wing.  In case the camera mount is different, but yaw, pitch, roll angles apply to the aircraft as described, it is necessary to use the GPS/INS offset values in the  Camera Calibration window.

Note that gimbal angles are not supported (varying angle offset for each camera position), so in such case pre-processing correction of the orientation angles should be used. Or angles should be ignored and not loaded to Reference pane.

As for the issue with the stereopair itself, there could be several reasons, for example, very short stereobase distance compared to the distance to the area of interest.
If you know the camera calibration parameters precisely, you can input them to the Camera Calibration window and use Fix Calibration flag to disable automatic distortion parameters refinement.

[PAOBC]

Hi Alexey. Thanks for the input. Below is more input per your answers.
GCPs - I actually have 51 markers (gcps) placed in each of the images that comprise the stereo pair. I've manually optimized the location of each marker in each image. I know these gcps are accurate because they were originated via triangulation w/ World View 1 reference imagery using Socet GXP. What is not clear to me is when to utilize the gcps. Do I have them initially unchecked when I perform the 1st "align photos"? Are the very large errors that I'm seeing in the Reference pane being driven by the gcps being checked "on" when I align the photos?

I'm still convinced that my issue lies w/ how the "Camera Position" parameters are defined. Per your 2nd paragraph the "camera center" would be equivalent to the platform center, meaning the lat/lon/altitude would be the geographic position of the platform. You note that GPS/INS offset values should be used. When you say "offset" do you mean the calibrated lever arm of the location of the GPS relative to the camera?

In your 3rd paragraph regarding the lens orientation angles (oblique look angle) you recommend either not using the lens mechanical pitch & roll angles OR applying a "pre-processing correction". What exactly would that be? I would think that the lens pointing orientation would be necessary for an accurate calculation of the actual image (lens) line of sight intersecting w/ the ellipsoid.

Given your recommendations I have already worked thru these various scenarios. Ironically the scenario where all the camera position parameters were platform specific (camera lens orientation ignored) resulted in the largest angular error in the Reference panel results.

Per your last comment regarding stereo b/h ratio, yes my camera has a very small ratio of ~0.15 & a convergence angle = 12 deg. That being said, stereo pairs from my camera fuse extremely well & are visually spectacular as obliquity increases.

Lastly, does PS utilize the 4-corner coordinates embedded in the GeoTIFF image? I see no reported geos in the Photos panel.

Thanks for your patience. I'm hoping we can come to a conclusion on how I can proceed toward my end goal or determine that my camera ops parameters is not suited for analysis w/ PhotoScan.