The accuracy gap between the high levels of accuracy in radiotherapy planning and the uncertain set-up of each therapy fraction represents a crucial factor in the optimisation of radiation treatment. This occurs because the conventional means of patient alignment and immobilisation do not guarantee accurate implementation of the therapy plan in the actual irradiation treatment. A patient repositioning technique is proposed, based on opto-electronic motion capture and on methods of registration of body surfaces described by a limited dataset. The validation of the method was related to breast cancer radiotherapy and was based on simulated and experimental repositioning procedures involving a phantom and two subjects. With respect to previous work, the surface registration procedure was, in this case, implemented as a constrained non-linear least-square problem (constraints were given by the position of a couple of passive markers placed on the sternum), and three different algorithms were compared in terms of accuracy in misalignment detection and of computational cost. The simulation and experimental activities identified the best performing algorithm, which systematically limited the repositioning errors to below clinically acceptable thresholds (5 mm), with residual surface mismatches lower than 2 mm.