The knowledge of the actual vehicle speeds is an essential characteristic of drivers behavior and their road usage. This information become available with the generalization of connected vehicles, but also smartphones, which increase the number of "tracers" likely to refer their position and speed in real time. In this thesis, we propose to use these digital traces and to develop a methodology, based on a functional approach, to produce several reference speed profiles. In a first part, we propose a functional modeling of space-speed profiles (i.e. speed vs position) and we study their properties (continuity, differentiability). In a second part, we propose a methodology to construct an estimator of a space speed profile from noisy measurements of position and speed, based on smoothing splines and the theory of reproducing kernel Hilbert spaces (RKHS). The third part is devoted to the construction of several aggregated profiles (average, median). In particular, we propose a landmark-based registration of profiles at stops, and we propose the construction of speed corridors reflecting the dispersion of actual speeds.
