D61 3D multiparticle tracking

Topic

This 3D multi-particle tracking method reconstructs the final trajectories of individual insects within a large swarm simultaneously, enabling the derivation of collective behavior statistics that would be impossible to obtain through the isolated tracking of a single insect.

The technical challenge of this tracking process is twofold: first, the 3D reconstruction of each particle from superimposed 2D images; and second, the correct assignment—from one frame to the next—of a specific detection to a particular individual within the swarm.

The Hungarian algorithm—an optimal assignment method—minimizes the total cost of matching detections to individuals, typically defined as the distance between an individual's predicted and observed positions. As the insect swarm becomes denser, the overlap between individuals becomes increasingly difficult to resolve, necessitating the use of additional cameras.

The swarm's interaction radius is a key parameter in this tracking process: a small radius results in few detected neighbors and limited statistical power regarding the swarm, whereas a large radius includes individuals that do not actually interact physically with one another within the swarm. Fine-tuning the swarm's interaction radius to its specific optimal value renders the resulting statistics on the swarm's collective behavior even more representative.

Conversely, the swarm's local density varies throughout the swarm itself, proving considerably higher toward the center and considerably lower toward the periphery.

Consequently, in this model—where the trajectory of each individual within the swarm is reconstructed simultaneously with the initial capture—the final three-dimensional uncertainty regarding each trajectory propagates from both the final two-dimensional localization error of each particle and the final geometry of the camera system itself. Improvements to the two-dimensional localization of each particle—achieved through the multi-particle tracking camera system—result in a further reduction of the final three-dimensional uncertainty regarding each swarm trajectory.

ID:2598