Aggressive, Tense or Shy? Identifying Personality Traits from Crowd Videos

Bera, Aniket; Randhavane, Tanmay; Manocha, Dinesh · 2017 · OpenAlex-citations

DOI: 10.24963/ijcai.2017/17

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of automatically classifying the dynamic behavior and personality traits of individual pedestrians within crowd videos. Motivated by the need for realistic crowd simulation in robotics, surveillance, and architectural design, the authors aim to move beyond generic crowd modeling by identifying heterogeneous individual behaviors. The research is grounded in Personality Trait Theory, specifically Eysenck’s PEN model, which posits that behavior variations stem from underlying traits. The authors propose a real-time algorithm that extracts pedestrian trajectories to compute local motion models and global crowd characteristics, enabling both personality classification and future crowd movement prediction. The methodology combines Bayesian learning with pedestrian dynamics techniques. The system processes streaming video input to extract 2D trajectories, using an Ensemble Kalman Filter and Expectation Maximization to estimate pedestrian states and compensate for noisy data. Local characteristics are derived from a Reciprocal Velocity Obstacle (RVO) motion model, mapping five parameters (e.g., neighbor distance, planning horizon) to six weighted behavior classes: aggressive, assertive, shy, active, tense, and impulsive. Global characteristics, including movement flows and start points, are learned using k-means clustering and Gaussian mixture models. These local and global features are integrated to predict crowd behavior under varying environmental scenarios. The authors evaluated the algorithm’s performance on diverse datasets ranging from low-density scenes to high-density crowds, including footage from the 2017 Presidential Inauguration. The system demonstrated robust real-time performance, processing videos with up to 200 pedestrians. To assess accuracy, the authors conducted a user study with 31 participants who labeled pedestrian behaviors in video clips. The algorithm’s predictions were compared against human judgments. Results showed that the algorithm correctly identified the most dominant personality trait in 76.96% of cases. When accounting for the second most dominant trait, the accuracy increased to 88.48%. The study also found that mapping the six specific traits to the broader three-factor PEN model improved agreement between algorithmic outputs and human perception. The significance of this work lies in being the first approach to automatically identify individual pedestrian personalities from video movements. By linking motion dynamics to psychological traits, the method enables more realistic and heterogeneous crowd simulations. The ability to predict crowd distribution and behavior based on learned personality traits offers practical applications for collision-free navigation of autonomous agents and improved crowd management in dense environments. The approach is generalizable to both indoor and outdoor settings without assumptions regarding crowd size or density.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• personality driving