Individual Footprint Recovery in Dense, Entangled Crowds

Enable reliable recovery of individual spatial footprints in dense, dynamic, and interaction-rich crowds where pedestrians merge, separate, and block one another, causing prolonged periods of highly entangled mmWave radar observations.

Background

The paper identifies individual spatial-footprint recovery as a difficult sensing and inference problem in dense pedestrian environments. Pedestrians interact, move in coupled groups, merge and separate, and can block one another, causing radar returns to become ambiguous or disappear for extended periods. These conditions create spatio-temporal entanglement in which observations cannot be straightforwardly assigned to individual people.

The stated challenge motivates the paper’s macro-to-micro framework, which combines learned crowd spatial semantics, a physics-informed model of radar observability loss, and semantic-guided multi-hypothesis reasoning. The paper presents this framework as a solution to the challenge, but explicitly characterizes reliable individual-footprint recovery under these dense, interaction-rich conditions as an open challenge.