Relative accuracy means you can only trust a measurement within a data set. That measurement cannot be trusted beyond that data set. In practical terms, with relative accuracy, you can trust something to be 50cm long +/- 2cm, but you cannot trust where that measurement is in the real world.
To achieve relative accuracy in a data set, all you need is a reference to scale. This can be an object in the area of interest whose dimensions are known, a purpose made scale bar or stereo pairs.