Socially Constrained Structural Learning for Groups Detection in Crowd

TL;DR

This paper introduces a novel social group detection algorithm in crowds using correlation clustering on trajectories, leveraging a learned affinity measure based on social and physical features, achieving state-of-the-art results.

Contribution

The work presents a new algorithm that combines social theories with structural SVMs for improved group detection in crowds.

Findings

Achieves state-of-the-art accuracy on trajectory data.

Effectively incorporates social and physical features.

Handles complex performance evaluation with a new loss function.

Abstract

Modern crowd theories agree that collective behavior is the result of the underlying interactions among small groups of individuals. In this work, we propose a novel algorithm for detecting social groups in crowds by means of a Correlation Clustering procedure on people trajectories. The affinity between crowd members is learned through an online formulation of the Structural SVM framework and a set of specifically designed features characterizing both their physical and social identity, inspired by Proxemic theory, Granger causality, DTW and Heat-maps. To adhere to sociological observations, we introduce a loss function (G-MITRE) able to deal with the complexity of evaluating group detection performances. We show our algorithm achieves state-of-the-art results when relying on both ground truth trajectories and tracklets previously extracted by available detector/tracker systems.