Abstract

We present a novel approach for analyzing the quality of multi-agent crowd simulation algorithms. Our approach is data-driven, taking as input a set of user-defined metrics and reference training data, either synthetic or from video footage of real crowds. Given a simulation, we formulate the crowd analysis problem as an anomaly detection problem and exploit state-of-the-art outlier detection algorithms to address it. To that end, we introduce a new framework for the visual analysis of crowd simulations. Our framework allows us to capture potentially erroneous behaviors on a per-agent basis either by automatically detecting outliers based on individual evaluation metrics or by accounting for multiple evaluation criteria in a principled fashion using Principle Component Analysis and the notion of Pareto Optimality. We discuss optimizations necessary to allow real-time performance on large datasets and demonstrate the applicability of our framework through the analysis of simulations created by several widely-used methods, including a simulation from a commercial game.

Video

Data Crowd Data

These are some of the datasets used in this work. The data consist of real world data from a commercial street (Zara.csv, Zara_cutoff.csv), filtered data from a bottleneck scenario (Bottleneck_2.5m.csv) and some data from a commercial video game (AssasinsCreed_1.csv, AssasinsCreed_2.csv). A single Comma-Separated Values file (.csv) contains spatio-temporal information of all characters present in the respective scene. The Bottlenect_obs.txt file contains the description of the obstacles in the bottleneck scenario.

Description

  • Column 1: The id of the character
  • Columns 2, 3: The (x, y) coordinates of the agent's position on the 2D plane it is moving.
  • Columns 4, 5: The direction (dir_x, dir_y) the character is facing. This can be different to the moving direction.
  • Column 6: The agent's radius. This typically has the same value during the life of an agent.
  • Column 7: The time t where the spatio-temporal data recorded at the current line were observed.

Example

id,x,y,dir_x,dir_y,radius,time
0,-1.71,2.14,0.496139,-0.868243,0.1,0
0,-1.7036,2.1288,0.496139,-0.868243,0.1,0.0333333
...
18,4.57,-0.39,-0.910369,0.413798,0.1,41.6666
18,4.5475,-0.3975,-0.919947,0.231291,0.1,41.7

Zara  Assassin's Creed 1  Bottleneck 

Citation

Please use the following when citing this work.

@article {charalambous2014data-driven,
    author = {Charalambous, Panayiotis and Karamouzas, Ioannis and Guy, Stephen J. and Chrysanthou, Yiorgos},
    title = {A Data-Driven Framework for Visual Crowd Analysis},
    journal = {Computer Graphics Forum},
    volume = {33},
    number = {7},
    issn = {1467-8659},
    url = {http://dx.doi.org/10.1111/cgf.12472},
    doi = {10.1111/cgf.12472},
    pages = {41--50},
    year = {2014},
}
University of Cyprus University of Minnesota          Research Promotion Foundation Republic of Cyprus European Union European Structural Funds



loading