How are RPCs from two sensors used to find the epipolar geometry to extract a DEM?

Question 1: In ENVI's DEM Extraction workflow, if you use a stereo pair that are not of the same sensor (SPOT and QuickBird for example), how are the RPCs (rational polynomial coefficients) from both images used together to extract the relationship between the stereo pair and elevation?

Question 2: If you use two images from different sensors, is there a minimum amount of overlap necessary to get a good DEM extraction?

Answers 1:
For a stereo pair that are not of the same sensor, the RPCs from both images will be used. The two sets of RPCs are used to extract elevation through RPC intersecting, which may be described as follows:

RPC1: x1 = f1(X,Y,Z)
y1 = g1(X,Y,Z)
RPC2: x2 = f2(X,Y,Z)
y2 = g2(X,Y,Z)

As you see, we have 4 equations and we have three unknowns(X,Y,Z), so we can solve the equation using the least squares method. Given ground coordinates (X, Y, Z), we can compute x1,y1,x2,y2 with RPCs, so we can figure out the relationship between the two images.

Answer 2:
There is not a minimum amount of overlap necessary to get a good extraction (unlike the case of classifical photogrammtry where 60% overlap is desirable). However, there is a requirement which states that two images should have a good intersecting condition. For example, if two images had 100% overlap, obtained with the same satellite position and attitude and viewing angle, then they are not a stereopair and cannot be used to extract a DEM or 3D measurement, because the intersecting condition is poor (their RPCs are the same).

We typically use BH ratio (Baseline to Height) to evaluate whether two images can be considered a good stereopair. In a simple but inaccurate term, baseline is the distance between two satellite positions when the images are captured, and the height is the flight height. The closer the BH ratio is to 1, the better the intersecting condition, and the images can be considered an acceptable stereopair for DEM Extraction. A BH ratio between 0.6-1.7 should be acceptable.

Review on 12/31/2013 MM