We propose an evaluation protocol to benchmark text-to-video (T2V) and image-to-video (I2V) models as implicit simulators of pedestrian dynamics. We use 3D reconstruction and depth estimation to extract pedestrian trajectories without known camera parameters.
We investigate whether image-to-video (I2V) models based on diffusion transformers can generate realistic pedestrian movement patterns in crowded public scenes by conditioning on keyframes from pedestrian benchmark datasets.
We describe an intelligent environment that unifies wearable skin-strain sensors for physiological monitoring, geophones and microphones to record ambient vibrations and sounds, and video cameras to visually observe human activities.