Modelling crowd motion is central to situations as diverse as risk prevention in mass events and visual effects rendering in the motion picture industry. The difficulty to perform quantitative measurements in model experiments, and the lack of reference experimental system, have however strongly limited our ability to model and control pedestrian flows. The aim of this thesis is to strengthen our understanding of human crowds, following two distinct approaches.First, we designed a numerical model to study the lane formation process among bidirectional flows of motile particles. We first evidenced the existence of two distinct phases: one fully laned and one homogeneously mixed, separated by a critical phase transition, unique to active syst...