Crowd editing

Using data assimilation techniques, we propose a method to efficiently mix crowd simulations and user-defined constraints. The model used for simulation has to be derivable. Constraints must be applied along with a corresponding observation operator. The later is in charge to sense the state of the crowd. Comparison to the constraint yields new values of the crowd model parameters, then new simulation, new crowd state and eventually new crowd sensing. We obtain a control loop producing iteratively the optimal crowd scene considering a given crowd model, initialization, and constraints.

We developed several observation operators. Each one of them corresponds to a specific constraint the user might need to edit a crowd scene. There exists two major classes of operator, for which was built different sub-classes for each of them:

Lagrangian

specific pedestrians positions, inter-pedestrians distances*

Eulerian

crowd density, velocity, divergence, vorticity

Related references:

1. Taesoo Kwon, Kang Hoon Lee, Jehee Lee, and Shigeo Takahashi. Group motion editing. ACM Trans. Graph., 27(3), 2008.
2. Antoine McNamara, Adrien Treuille, Zoran Popovic, and Jos Stam. Fluid control using the adjoint method. ACM Trans. Graph, 23(3):449--456, 2004.

* With joint work on particle swarm control.

Agoraset

We present a simulation-based crowd video dataset to be used for evaluation of crowd analysis methods, such as tracking or segmentation. Most of the time, an exact ground truth associated to real videos is difficult and time-consuming to produce, prone to errors, and these difficulties rise exponentially with the apparent density of the crowd in the image. We propose a synthetic crowd dataset to help researchers evaluate their methods against a temporally dense synthetic ground truth.

Without Agoraset

With Agoraset

Several rendering methods of the synthetic sequences are proposed, such as shadows or sky luminosity rendering. Disposing of a same scene with realistic visual effects, and without, allows measurement of the perturbation of the analysis method due to luminosity conditions.

Ground truth examples

Here are a some examples of what can be generated using the dataset ground truth. These materials are made easily with matlab routines also furnished in Agoraset's website.

Related references:

1. Mikel Rodriguez, Josef Sivic, Ivan Laptev, and Jean-Yves Audibert. Data-driven crowd analysis in videos. In Proceedings of the 2011 international conference on computer vision, ICCV '11, 1235--1242, Washington, DC, USA, 2011. , IEEE Computer Society.
2. Simon Baker, Daniel Scharstein, J. P. Lewis, Stefan Roth, Michael J. Black, and Richard Szeliski. A Database and Evaluation Methodology for Optical Flow. International Journal of Computer Vision, 92(1):1--31, novembre 2010.
3. D. Helbing, I. Farkas, and T. Vicsek. Simulating dynamical features of escape panic. Nature, 407(1):487--490, 2000.

Crowd flow characterization

Using a simple continuous crowd model based on velocity and density, and by applying variational assimilation techniques using adjoint methods, we propose a method to retrieve disturbance and density distribution on crowd video footage.

Related references:

1. E.L. Andrade, S. Blunsden, and R.B. Fisher. Modelling crowd scenes for event detection. In Pattern recognition, 2006. icpr 2006. 18th international conference on, volume 1, 175-178, , , 2006.
2. D. Helbing, I. Farkas, and T. Vicsek. Simulating dynamical features of escape panic. Nature, 407(1):487--490, 2000.
3. François-Xavier Le Dimet and Olivier Talagrand. Variational algorithms for analysis and assimilation of meteorological observations: theoretical aspects. Tellus A, 38A(2):97--110, 1986.
4. J.L. Lions. Optimal control of systems governed by partial differential equations. Springer-Verlag, 1971.