Crowd Density Analysis with Marked Point Processes [Applications Corner]

This article presents a Bayesian approach that estimates the count and location of individuals in a video frame. Crowds are modeled by a marked point process (MPP) that couples a spatial stochastic process governing number and placement of individuals with a conditional mark process for selecting body size, shape, and orientation. Given a noisy, binary mask image where pixels are labeled foreground or background, the approach seeks a configuration of cutout shapes that simultaneously "covers" as many foreground pixels and as few background pixels as possible.