D10 Computer Vision of Formations

Topic

A synchronized multi-camera system, combined with computer vision algorithms, enables the reconstruction of the three-dimensional movement of an entire flock—bird by bird—rather than merely describing the group's average behavior. This 3D reconstruction of flock movement allows for the direct observation of both individual aerodynamic dynamics and the collective behavior of the flight formation, as well as the anomalies that arise when that formation breaks apart due to an external disturbance.

The central technical challenge of this synchronized multi-camera system is tracking each individual bird within the flock; a scene featuring dozens or even hundreds of rapidly moving birds—many of which are momentarily obscured by others—requires computer vision algorithms capable of tracking each bird's identity from one frame to the next without confusing it with its neighbors.

Current systems already address this technical challenge on a significant scale: the system developed by Cavagna and his colleagues—which combines ten synchronized cameras capturing 170 frames per second—successfully resolves the individual trajectories of every bird in flocks of up to two thousand starlings.

Based on these individual trajectories, the method calculates all the collective variables required by the simulator. The first is the distance between neighbors—the separation each bird maintains relative to the nearest birds in the flock at any given moment—a metric that allows for the direct validation of how each bird adjusts its position in relation to its neighbors within the collective flight formation.

The second collective variable is the Vicsek order parameter itself—a metric summarizing the degree of alignment among the flight directions of all birds in the flock at a given instant. Calculated directly from the individual trajectories provided by this method, the Vicsek order parameter describes the global state of the collective flight formation at any moment, eliminating the need for indirect inference. The third collective variable is the V-shaped geometry formed by certain flocks, such as those of geese; by directly measuring the relative position of each bird with respect to the others within that V, the method makes it possible to verify whether each bird truly occupies the optimal position relative to the bird ahead—the position that allows it to benefit from the vorticity field generated by that leading bird during flight.

The fourth collective variable is the speed at which disturbances propagate through the flock; by measuring the time it takes for a disturbance affecting one bird to reach its neighbors—and subsequently the neighbors of those neighbors—the method calculates this propagation speed directly from individual trajectories. In the case of starlings, this propagation speed exceeds the individual flight speed of the birds themselves. This propagation speed serves as a key diagnostic indicator of the collective system's robustness in the face of a disruption to the formation: the faster a disturbance propagates through the flock, the more quickly the flock can reorganize to restore its order.

By combining inter-neighbor distance, the Vicsek order parameter, V-shaped geometry, and disturbance propagation speed, this three-dimensional reconstruction of flock movement—derived entirely from the individual tracking of each bird via computer vision—allows for the direct and unambiguous validation of all collective variables describing both an intact flight formation and one recovering from a disruption.

ID:4230