Comparison of spatial partitioning data structures in crowd simulations

Abstract: This report investigates how the construction and query time of multiple spatial partitioning data structures is impacted by spatial distribution of and number of agents in a crowd simulation. In addition a method is investigated for updating the data structures less frequently at the cost of increasing the radius queried, without affecting the correctness of the queries. The data structures are tested in a simulation using a Boids model and update and query times are measured. It is found that the performance of the grid is better than the quad tree and the kd- tree for low number of agents, but deteriorates more quickly when the number of agents increase. It is also found that this approach can decrease the sum of time spent updating and the time spent querying in the simulation. The effectiveness of this method is highly dependent on the update of the data structure.