We operate at the interface between complex flowing matter, physics for society and machine learning for non-linear physical systems. Our goal is to advance our fundamental understanding of pedestrian crowd flows. We aim at quantitative models for the emergent physics of crowds to allow safer and more efficient pedestrian environments. To this purpose, in collaboration with national and international facility managers of, e.g.,  municipalities, museums, and festivals, we employ large-scale real-life crowd tracking experiments. Our activity includes also the application of recent machine and deep learning techniques to the analysis of highly complex and non-linear physical systems and, in particular, fluid turbulence.