Crowd Detection with a Multiview Sampler

Abstract

Abstract. We present a Bayesian approach for simultaneously estimating the number of people in a crowd and their spatial locations by sampling from a posterior distribution over crowd configurations. Although this framework can be naturally extended from single to multiview detection, we show that the naive extension leads to an inefficient sampler that is easily trapped in local modes. We therefore develop a set of novel proposals that leverage multiview geometry to propose global moves that jump more efficiently between modes of the posterior distribution. We also develop a statistical model of crowd configurations that can handle dependencies among people and while not requiring discretization of their spatial locations. We quantitatively evaluate our algorithm on a publicly available benchmark dataset with different crowd densities and environmental conditions, and show that our approach outperforms other state-of-the-art methods for detecting and counting people in crowds.

Crowd Counting and Profiling: Methodology and Evaluation

“…Distribution free decomposition of multivariate data. Pattern Analysis and Applications, 2(1): [22][23][24][25][26][27][28][29][30]1999.…”

Automatic Crowd Analysis From Very High Resolution Satellite Images

“…It requires clear detection of body boundaries, which is not possible in airborne images. In another study, Ge and Collins (Ge and Collins, 2010) used multiple close-range images which are taken at the same time from different viewing angles. They used threedimensional heights of the objects to detect people on streets.…”

Multi-view people surveillance using 3D information

“…Secondly, due to occlusion, pixels corresponding to different objects can be merged in the same connected blobs of the motion masks. To handle the above challenges, multi-view approaches [8,10,19] have recently been proposed. The method in [8] uses a discretized grid on the ground plane, and assumes that the people have approximately a uniform height.…”

“…The method in [8] uses a discretized grid on the ground plane, and assumes that the people have approximately a uniform height. [10] attempts to obtain a configuration which explains the observed data with a minimal number of occlusions, expecting that people should not be occluded in all views. Both methods [8,10] attempt to match the complete projections of the proposed object silhouettes to the observed foreground masks, thus they strongly depend on the quality of the background subtraction step.…”

“…[10] attempts to obtain a configuration which explains the observed data with a minimal number of occlusions, expecting that people should not be occluded in all views. Both methods [8,10] attempt to match the complete projections of the proposed object silhouettes to the observed foreground masks, thus they strongly depend on the quality of the background subtraction step. Similarly to [19], we purely focus on the head and leg regions, when we calculate simple pixel-level features from the projections of foreground pixels on multiple parallel planes.…”