Formation sketching: an approach to stylize groups in crowd simulation

Most of existing crowd simulation algorithms focus on the moving trajectories of individual agents, while collective group formations are often roughly learned from video examples or manually specified via various hard constraints (e.g., pre-defined keyframes of exact agent distributions). In this paper, we present an intuitive yet efficient approach to generate arbitrary and precise group formations by sketching formation boundaries. Our approach can automatically compute the desired position of each agent in the target formation and generate the agent correspondences between keyframes. When high-level group formations need to be formed on-the-fly in a dynamic environment such as "switching to the circle formation at about one hundred meters ahead", our algorithm will coordinate and compute appropriate actions for each agent by seamlessly fusing local formation dynamics and global group locomotion. Through a number of experiments, we demonstrate that our approach is efficient and adaptive to variations of group scales (i.e., number of agents), group positions, and environment obstacles.