Image tiling constrained by navigation RRS feed

  • Question

  • I'm an oceangrapher interested to see if ICE can help in mosaicing underwater images collected from an underwater robot. I've tried it and indeed, for short sequences of images when the seafloor is flat, it does a fantastic job. But for longer sequences, even when I attempt a "structured" panarama with planar motion, ICE tends to get confused and produces a less than desirable result.

    The thing is, I have lots of information about each image, including its approximate position, orientation and the average distance to the seafloor. None of this information is "pixel level" accurate so I can't use it to directly mosaic the images, but it certainly constrains the problem greatly. So I am asking for suggestions from the community and maybe a feature request.  

    I'd like to be able to load a text file along with a sequence of images. The text file would have filenames, and optinally estimated position (in arbitrary or geographic units), orientation of the field of view (pitch, roll, heading) and maybe estimated mean distance of objects in the image and have these used to constrain the resulting panaorama.

    For example, if an underwater robot takes a series of images of the seafloor along a straight line, if the seafloor slopes to one side the resulting mosaic will curve, as one cannot tell from the images alone that the surface is sloping. But if it was known that the images were taken along a straight line, then the mosaic could be generated keeping this constraint in mind, the resulting mosaic would be linear and objects in the mosaic might be sized properly relative to each other.

    I hope all this makes sense and I hope the folks behind ICE will agree that this is a hugely useful addition that has many applications beyond my own, and will code it up!



    Wednesday, June 6, 2012 4:39 PM