Publication (Eurographics 2019): Character navigation in dynamic environments based on optical flow


The objective of microscopic crowd simulation algorithms is to calculate the motion of a large number of agents.  The movement of agents results from a set of control laws that allow them to perform actions in their environment, such as reaching a destination, or to perform interactions with other agents, such as collision avoidance.

In general, the control laws necessary for these actions are generally formulated on a geometric basis. For example, moving towards a destination may involve reducing the angle between the target direction and the direction of movement, while progressing to reduce the distance to the target. These geometric relationships are simple to transform into computer code for simulation. Good,…

However, this geometric approach takes us away from the way humans perform these movements, because they do not directly solve geometry problems. They are guided because they perceive, and in particular by what they see of their environment. This process by which a human being controls his movement from what he sees is called a “perception-action loop”, and in our particular case a “visuo-motor” loop.

The objective of Axel Lopez’s work, a doctoral student in our team, is to simulate the visuo-motor loop that guides human locomotion in a dynamic context. This work is a departure from previous approaches based on a geometric formulation. In particular, it implicitly takes into account some of the perceptual factors that influence human navigation: the visibility of objects, the saliency of their relative movement, etc.

One limitation of such approach is however the computational cost of simulating the visual perception. We expect to work on these aspects in the near future, and to improve the performances of our technique so as to make it applicable to crowd simulation.



López, A., Chaumette, F., Marchand, E., & Pettré, J. (2019, May). Character navigation in dynamic environments based on optical flow. In Computer Graphics Forum (Vol. 38, No. 2).

Comments are closed.