Elouan GROS

An Overview of Herd Behavior Synthesis

Elements at our disposal

To produce a convincing synthesis, we collected some data in previous steps. Firstly, we collected a configuration space from video data, in order to learn actual herd behavior and configuration. This data contains speed, density and neighbor information for the sheep detected in the video. Then, we collected user input, to determine the path along which our synthetic herd was to move along, in the form of a direction field. This information takes the form of a vector field, where every vector is a unit vector that determines where the sheep should orient itself Finally, we used the video data to determine the changes in orientation of the sheep in function of the density around them. This information is encoded in a change of orientation function, that we can use to generate changes in orientation.

First intuition

Our first intuition is to use the following relation:

Vn+1 = S(dn, |Vn|) * ρ( G( μ = 0, σ = θ( dn ) ) ) * Ο[ cn ] + F

Where, for a given animal:

• V_n is its velocity at frame n
• |V_n| is its speed at frame n
• d_n is its density at frame n
• c_n is its position at frame n

And:

• S(d,s) is the function that to a density d and speed s, maps the speed of the animal from the configuration space that most closely matches d and s
• ρ is a function that to an angle maps its change of rotation matrix
• G is a Gaussian distribution of parameters (μ, σ)
• θ is the change of orientation function
• Ο[x] is the value of the direction field at position x
• F is the sum of some forces we will explain later