QUOTE (djellison @ Jul 9 2012, 01:59 PM)
I would doubt, very much, that it would be visible.
The trick here is that you can take advantage of measurements along each edge and diagonal of the base, then do pairwise comparisons between these measurements on different dates. So it is not so critical to observe the change on a single edge -- let's say you measure the 4 edge lengths and the 2 diagonals, then scale the diagonal lengths to match the edges (divide by sqrt(2)), so you've essentially got 6 samples of the edge length of this square base on that occasion. Do the same thing on a different occasion. Perform a paired t-test by pairing the equivalent measurements from the 2 occasions. From this you've got a mean change in the edge length, and you can get an associated confidence interval and p-value. The "noise" in the measurements is essentially averaged out, and since you can quantify the noise (since you can get a standard deviation) you can do some good statistics; handily, this sort of measurement error should be very "normal".
Paolo, the pointing of the pancam need only be such that the target is included in both L & R images taken on that occasion, the exact orientation of the pancam is not critical, we're depending on the camera model to identify the xyz coords of each vertex on the plate, then finding the 3D distances between those vertices. Where this gets potentially complicated is if you stop to consider whether the change in temperature also changes the camera model enough to be a problem.
My back-of-envelope estimates at the time I first looked at this suggest that a 100 degree C change in temperature should be easily measurable, and likely even a 20 degree change. Below that it looked "iffy".