Incorrect Projection of Horizontal Images - Flipped Vertically?

[rocdoctor]

I am using Photoscan 1.2.3 and have a number of images taken of a rock face from a horizontal camera position.  When I import the images into Photoscan, with image locations taken from the EXIF data, the model defaults to a vertical view of the model, rather than positioning the cameras horizontally and a view in the horizontal plane.  This results in the model being incorrectly orientated in real-world space with respect to the x, y, z axes.  The projection seems to assume the photos have been taken from a vertical rather than horizontal position.

When I export this data and use it to make measurements of the orientation of features on the rock face, all the orientations are completely out with respect to north.  If I rotate the image to the correct plane of view, i.e. front on to the face, none of the x,y,z axes are orientated correctly with respect to north.  For information, the view is taken looking towards the west, so that north (y-axis) should be orientated to the right.  I would like to export the point cloud as a properly georeferenced model.

If I clear the EXIF data from the model, I can of course rotate the object and redefine the region, so that the model is approximately correctly orientated, but this seem a very imprecise way of doing it.  Also, as soon a the model is georeferenced it is reprojected to the top-down view

Is there something incorrect with the way the images have been taken, or is this an issue with Photoscan?  I attach a couple of images to illustrate my point. 

I realise that some of the images share coordinates, as they were taken  from the same position by panning up or down, though there are 4 separate sets of coordinates.  Do I need to take the photos from more positions to get the alignment?

[Alexey Pasumansky]

Hello rocdoctor,

I suggest that you need better distribution of the reference points (camera position coordinates in this case) in space, as it seems that there are only four points that lay almost on the same line.

Also taking images from the same position is not good due to no stereo base distance provided in such case, unless you are using nodal head for image capture, but in this case the corresponding images should be grouped into Camera Station folders in the Workspace pane.

[rocdoctor]

Thank you for your reply Alexey.  I thought it might be the case that there weren't enough camera position in these images.  They were originally taken for conventional stereopair photogrammetry.  Also some good advice for using nodal heads for image capture. 

[Alexey Pasumansky]

Hello rocdoctor,

Having control points in scene with the measured coordinates may be also helpful for the valid coordinate system definition.