Not completely. If the scene is truly planar, or effectively flat because the subject is far away, you can often align the images with a homography after calibrating the cameras. In that case you don’t need per-object distance.

For a general 3D scene, parallax depends on depth, so you can’t avoid distance entirely. Using two visible cameras can help: you can estimate depth with stereo matching, then render a virtual visible view from the thermal camera’s position. But that is still using distance information indirectly via a depth map, and it can fail in occluded or low-texture areas.

A practical alternative is to reduce parallax physically by mounting the cameras as close together as possible. A beam-splitter setup can place two optical paths nearly at the same viewpoint, which greatly simplifies registration. You would still need a one-time calibration for focal length, distortion, and alignment.

So: flat scene → homography may be enough; general 3D scene → you need stereo/depth or a near-coincident optical setup.