Coordinate system orentation error single pass

[mapper]

When I import a single pass of imagery into photoscan it doesn't set the coordinate system with the z-axis up. Is there some trick to import the imagery that enables the correct import? I have tired imagery with the GPS position information in the EXIF headers, I have tired the GPS as a separate file both with and without INS (pitch, roll and yaw) data. The imagery stitches and scale well, but the model is effectively rotated about the long axis a seemingly random number of degrees between 20-45 out of Nadir. I would appreciate any help/pointers.

[Alexey Pasumansky]

Hello mapper,

Could you please post a screenshot of the Ground Control pane screenshot and model overview?

[Alexey Pasumansky]

Also please note that yaw, pitch, roll angles are not used for referencing, so based on the camera position data it is not possible to georeference straight single flight line due to the rotational degree of freedom around the flight direction.

[mapper]

Alexey,

I am unclear as to why the orientation information would be ignored when processing the imagery? How do you moved from of the relative coordinate system to an earth centered earth fixed system without understanding the image orientation to at least calculate the boundaries of your model? I understand you could impute orientation data if you have big enough bundle block with sufficient overlap but why wouldn't you use them if provided? Feel free to point me to some other reference to answer this new question.

Let me make sure I understand your statement correctly, the models used are unable to determine image orientation about the axis of travel for a single pass. Is that correct? If so I believe we have run squarely into that situation. Attached are two screen shots of a small single pass model made from a nadir pointing color DSLR system. The area is a flat coastal area. I need to be able to get a correct orthomosaic of this data. At this point AgiSoft has pretty much all the positioning and scaling data correct except for the rotation about the x axis which effective means z which should align (for the most part) with the camera pointing directions isn't even close. I've done several imports and there seems to be no clearly set angle for the rotation about the x axis it is pretty much random.

If I understand your statement correctly there are probably only two possible solutions. Impute a rotation about x or use ground control points. At this point I don't have GCPs and would prefer not to go back and get them if possible so it it possible to impute the correct orientation for the data? I understand that this solution probably won't be as accurate as the GCPs but may be accurate enough.

Thanks,
Gabe