If you were going to try it, I'd use some panorama stitching software to align the pairs of images (align the thermal images to the DSLR images) and make the output image size the same. That would require manually selecting control points which sounds a bit tedious.
Make 2 folders of images, using matching file names for the pairs. Make a 3rd folder to store the images for access by Photoscan. Put the DSLR images into this and make your point cloud, then swap in you thermal images and make your texture (after building the mesh). If you want to colour the point cloud from the thermal images you can transfer the colour from the texture image in an external program like Meshlab.
Before that though, I'd try making models from both sets of images, add matching GCPs and align the chunks. Merge them and then disable the DSLR images to make the texture... success may vary depending on detail in the thermal images.