Detecting anomalies in dense 3D crowds

Melania Prieto-Martín, Marc Comino-Trinidad, Dan Casas

Abstract: Highlights•A new method for anomalies in velocity in dense 3D crowds.•A new pipeline to generate synthetic annotated data.•This pipeline can generate data from real or simulated crowd trajectories.•A new dataset of velocity anomalies based on optical flow, depth and segmentation.•Anomalies labeled via normal scenarios velocity likelihood.