A Spatio-Temporal Hierarchical Model for Crowd Formation Planning in Large-Scale Performance

Abstract

Crowd formation of aesthetic transformation is considered to have extremely high artistic value and is widely applied in large-scale performances. In this paper, a spatio-temporal hierarchical model that parts the crowd formation transform into multiple granularities is proposed. Its core idea is to add spatio-temporal constraints created by directors into transformation process after multi-level division. In this model, average hash value and energy optimization are used to achieve reasonable crowd formation arrangement, while smooth and collision-free formation transformations are presented by constrained region growth and Kuhn-Munkres algorithm. We have also proposed a framework to achieve the generation of visually pleasing crowd formation transform performance based on the constraints. Besides, a virtual crowd formation transformation simulation was built to verify the effect of the proposed model. Through simulation experiments and comparisons, it was demonstrated that this hierarchical model can generate aesthetic crowd formation transformation with a satisfactory process.

Anchor‐based crowd formation transformation

“…In addition to the global moving distance, researchers have proposed a number of indicators to measure crowd formations, such as stability of the local structure, 1 effort balancing 1 and fitness of appearance to the director's intention. 3 However, after interviews with several international directors, we concluded that artistic aesthetics are not entirely objective and cannot be qualitatively measured through metrics. This does not mean that metrics are completely meaningless.…”

“…Crowd formation transformation can be described as a series of processes that transform between crowd formations. The crowd formation can be expressed as a set of agents 3 , or other linear space. Moving from one crowd formation to another is called a transformation:…”

“…As shown in Figure 2, the traditional method divides the crowd formation transformation generation into three steps: (1) formation generation, (2) macroscopic direction, and (3) microscopic movement, in which process (2) achieves the macroscopic planning of transformation by overall or group matching. 1,3 The diversity of the transformation of the crowd formation also depends on the process (2). This study optimized the process as a loop iteration of (i) anchor selection, (ii) agent assignment, and (iii) simulation and feedback to obtain a crowd formation transformation with high diversity, high boundary retention, and low overall loss of motion.…”

“…Because of the introduction of spatio-temporal constraints on crowd motion, the crowd formation transformation has the characteristics of least total effort, 1 collision-free, 2 and shape-based. 3 Thus, crowd formation transformation presents different visual aesthetics than unrestrained crowd movements.…”

“…Current research is based on specifying a series of key processes that enable the crowd to transform from a source shape to a target shape through smooth, collision-free motion. [1][2][3][4] Many studies have addressed this process through bipartite graph matching optimization or constraint-based dynamic co-optimization methods. [5][6][7] Such approaches, however, have failed to address the following problems.…”