Although pedestrian crossings are supposed to guarantee safety, accidents are still frequent. Prior to implementing any measures to address this issue, the safety of the pedestrian crossing must be assessed. Various indices estimating the safety of pedestrian crossings have been developed. They quantify spatio-temporal distance, potential collision severity or kinetic energy on impact. Computing these indices requires accurate and reliable trajectory data which can be extracted from videos thanks to computer vision algorithms. Like any measurement technology, this process produces imperfections and uncertainty in its output. These errors and uncertainties are scarcely considered in the current index definitions. This project aims to integrate these different measurement uncertainties into the calculation of such indices. Instead of computing a single deterministic value for each index, we propose a methodology providing the index probability distribution under various uncertainties. Further- more, we investigate the influence of each uncertainty on the index distribution. The feasibility and reliability of this method are verified by applying it to an empirical data set. The methodology is validated by comparing the results to state-of-the-art conflict analysis. We observe that the outcome of the indices is strongly conditioned by the quality of the data, in particular the smoothness of the trajectories.