Crowd behavior representation: an attribute-based approach.

Hamidreza Rabiee,Javad Haddadnia,Hossein Mousavi

doi:10.1186/s40064-016-2786-0

Hamidreza Rabiee, Javad Haddadnia + Show 1 more

Open Access

https://doi.org/10.1186/s40064-016-2786-0

Copy DOI

Abstract

In crowd behavior studies, a model of crowd behavior needs to be trained using the information extracted from video sequences. Most of the previous methods are based on low-level visual features because there are only crowd behavior labels available as ground-truth information in crowd datasets. However, there is a huge semantic gap between low-level motion/appearance features and high-level concept of crowd behaviors. In this paper, we tackle the problem by introducing an attribute-based scheme. While similar strategies have been employed for action and object recognition, to the best of our knowledge, for the first time it is shown that the crowd emotions can be used as attributes for crowd behavior understanding. We explore the idea of training a set of emotion-based classifiers, which can subsequently be used to indicate the crowd motion. In this scheme, we collect a large dataset of video clips and provide them with both annotations of “crowd behaviors” and “crowd emotions”. We test the proposed emotion based crowd representation methods on our dataset. The obtained promising results demonstrate that the crowd emotions enable the construction of more descriptive models for crowd behaviors. We aim at publishing the dataset with the article, to be used as a benchmark for the communities.

Highlights

Learning-based methods for human behavior recognition have been the subject of various studies over the last years
Crowd behavior dataset after a brief review on the state-of-the art crowd datasets for the task of crowd behavior analysis, we present our dataset in details
Regarding emotion attributes as an abstract part of a behavior class, we introduce a semantic space En where in location of an emotion attribute is defined as a latent variable, ei ∊ En

Summary

Introduction

Learning-based methods for human behavior recognition have been the subject of various studies over the last years. (Chen et al 2007; Kratz and Nishino 2009, 2010; Krausz and Bauckhage 2011, 2012; Li et al 2014; Mahadevan et al 2010; Mehran et al 2009; Raghavendra et al 2011; Rodriguez et al 2011; Roggen et al 2011; Saxena et al 2008; Solmaz et al 2012; Su et al 2013; Wang et al 2012; Zhang et al 2012) These features are directly related to behavior types (such as panic, fight, neutral, etc.) using modern machine learning techniques, e.g. support vector machines. In Krausz and Bauckhage (2011, 2012), optical flow histograms are used to demonstrate the global motion in a crowded scene They derive the histogram of the optical flow and extract some statistics from it to model human behaviors. On the other hand, introduce models extracted from fluid dynamics or other physics laws to model a crowd as a group of moving particles

Methods

Findings

Conclusion