Crowd density map estimation based on feature tracks

Abstract

Crowd density analysis is a crucial component in visual surveillance mainly for security monitoring. This paper proposes a novel approach for crowd density measure, in which local information at pixel level substitutes a global crowd level or a number of people per-frame. The proposed approach consists of generating fully automatic and crowd density maps using local features as an observation of a probabilistic crowd function. It also involves a feature tracking step which allows excluding feature points belonging to the background. This process is favorable for the later density function estimation since the influence of features irrelevant to the underlying crowd density is removed. Our proposed approach is evaluated on videos from different datasets, and the results demonstrate the effectiveness of feature tracks for crowd estimation. Furthermore, we include a comparative study between different local features in order to investigate their discriminative power to the crowd.

Spatio-temporal crowd density model in a human detection and tracking framework

“…For this, generating locally accurate crowd density maps is more helpful than computing only an overall density [23] or a number of people [24] in a whole frame. In the following, our proposed approach for crowd density estimation [25] is presented. First, local features are extracted to infer the contents of each frame under analysis.…”

Crowd analysis in non-static cameras using feature tracking and multi-person density

“…We follow the paradigm of Fradi and Dugelay [7] where a relation between the density of local features in the foreground and the underlying crowd density is assumed. Accordingly, this model implies a general link between the number of foreground features and the number of persons in the image.…”

“…by thresholding for a minimal average motion as proposed in [7]. This has been shown to be very suitable but is only valid for static cameras.…”

“…A disadvantage is that due to this one-frame motion information only currently moving persons are considered and no camera motion is allowed. In another approach, Fradi and Dugelay [7] compute crowd density derived from a distribution of moving FAST features using kernel density estimation. The resulting density maps give a first impression of highly-crowded locations but do not represent the spatio-geometrical properties of groups and are not suitable for segmentation.…”

Spatial and temporal variations of feature tracks for crowd behavior analysis

“…We therefore resort to another crowd measure, in which local information at pixel level substitutes a global number of people or a crowd level by frame. The alternative solution [15] is indeed more appropriate as it enables both the detection and the location of potentially crowded areas.…”