EN FR
Homepage Inria website


Section: New Results

Virtual Human Simulation

In addition to this last contribution on biomechanically-inspired character simulation, at the crossroad between motion analysis and simulation, we also explored two main directions for virtual human simulation in 2016. Firstly, with the arrival of Antonio Mucherino in the team, we pushed the idea of extending the idea of interaction meshes (introduced in 2010 by Taku Komura in Edinburgh) to model the constraints intrinsically associated with the motion. This approach requires developing new distance geometry algorithms in order to take time and rigid body constraints into account. Secondly, we continued to push the idea of using perceptual studies to efficiently adapt simulation in order to save computation time for less important details.

Julien Pettré moved to the Lagadic Inria team in March 2016. However we continue collaborating with him on crowd simulation problems, e.g., developing models related to interactions between pedestrians and designing perceptual studies to improve the realism of simulations.

Recent advances in discretizable distance geometry

Participants : Antonio Mucherino, Ludovic Hoyet, Franck Multon.

Since September 2016, Antonio Mucherino has a half-time Inria detachment in the MimeTIC team, in order to collaborate on exploring distance geometry-based problems in representing and editing human motion. In collaboration with various French and international partners, he has been working on the different facets of the discretization of the distance geometry. In 2016, he has mainly focused on the two following points. Firstly, since the discretization assumptions require the existence of a vertex ordering on the graph G which is used for representing a problem instance, he presented a new algorithm for the automatic detection of vertex orders that are also able to optimize a given set of objectives [7]. With the aim of making its exploration more efficient, the idea is to reduce in size the search space obtained with the discretization, while keeping in its interior the entire solution set. Secondly, he has started to investigate the possibility to extend the distance geometry (and its discretization) to a wider range of applications, by studying the overlaps between two different geometrical applications, arising in two different domains [3].

More related to the integration with the work in MimeTIC, we are currently exploring applying distance geometry approaches to other applications of interest for virtual human simulations, such as human motion editing and retargeting, and crowd simulations.

Perception of Secondary Motions in Crowd Scenarios

Participants : Ludovic Hoyet, Anne-Hélène Olivier, Richard Kulpa, Julien Pettré.

Creating plausible virtual character animations is of importance in topics researched in MimeTIC, especially for interactive applications where balancing realism and computational load is a requisite. Recently, we investigated how to improve realism of virtual crowd animations by exploring the effects of introducing secondary shoulder motions at the animation level. Typically, a crowd engine pipeline animates numerous moving characters according to a two-step process. First, a crowd simulator generates the characters’ global 2D displacement trajectories in the environment, then an animation engine transforms these global trajectories into full body motions. This two-step decomposition is interesting for computational reasons, as crowd simulators raise quadratic complexity issues by nature. For the sake of simplicity, simulation models are often limited to 2D moving circles with 3 degrees of freedom (DoF), i.e., two translations and a rotation. The complete set of internal trajectories (30 to 60 DoF per character) is then considered at the animation step only, where characters are processed independently. This two-step process avoids combining the complexity of crowd simulators with the dimensionality of character kinematic models. However, it also leads to the notion of interactions between characters to be considered only at the simulation level, and to be lost at the animation level. Body animations are therefore not influenced by the presence of neighbours, only global trajectories are. Final animations therefore often lead to residual collisions and/or characters walking as if they were alone, showing no sign to the influence of others.

In this work, we investigated the value of adding secondary motions on the perceived visual quality of crowd animations (i.e., perceived residual collisions and animation naturalness). We focused on adding shoulder motions to characters passing at close distances, and explored this question through two perceptual experiments. To understand the effects of shoulder motions on walking interactions, we first focused on understanding how these secondary motions affect how viewers perceive local interactions between two characters. We found that shoulder motions have strong positive effects on the visual quality of two-character animations, where such animations are perceived to be significantly more natural, and residual collisions become significantly less perceptible. Then we evaluated the benefits of displaying shoulder motions in the situation of crowded scenes, where shoulder motions are diluted into much more visually complex animations, and demonstrated positive effects on the animation naturalness. This increase of visual quality is obtained at a very low computational overhead, which demonstrates the relevance of the direction explored by our work. Our general conclusion is that adding secondary motions in character interactions has a significant impact on the visual quality of crowd animations, with a very light impact on the computational cost of the whole animation pipeline. Our results advance crowd animation techniques by enhancing the simulation of complex interactions between crowd characters with simple secondary motion triggering techniques.

These results were accepted and presented in SIGGRAPH 2016, the premier and most selective computer graphics scientific event, and published in ACM Transaction on Graphics [11].