Understanding Avian Flocks Near and Entering Civil Infrastructure for Species of Concern

Project ID: W912HZ-17-2-0005

Federal Agency: U.S. Army Corps of Engineers

Partner Institution: Portland State University

Fiscal Year: 2017

Initial Funding: $30,510

Total Funding: $309,103

Project Type: Research

Project Disciplines: Biological

Principal Investigator: Siderius, Martin

Agreement Technical Representative: Noel, Mark

Abstract: This is the statement of work for a one-year study of the collective behavior of Vaux’s swift flocks. This one-year study is meant to establish scientific links between the Collective Behaviour in Biological Systems at ISC-CNR and Portland State University (PSU). There are several components of the project that need preliminary study in this one-year project.
In an effort to establish solid scientific links between the COBBS lab (Collective Behaviour in Biological Systems) at ISC-CNR and the NEAR-Lab (Northwest Electromagnetics and Acoustic Research) at PSU, Dr Melillo (CNR) and Dr Siderius (PSU) conducted a pilot season of singlecamera 2D data-taking in September 2016 on Vaux’s swift at Chapman Elementary School, Portland. Preliminary analysis performed at the CNR end of these images demonstrate the feasibility of a 3D experimental study of this system. However, the images also show that flocks of Vaux’s swifts at Chapman Elementary School perform a global rotational motion (around the chimney of the school) that is specific to Vaux’s swift and that was absent in data analyzed by CNR in the past on different species (starlings). This rotational motion may prevent current 3D tracking algorithms from performing efficiently, if not jeopardize their functionality altogether.
The essential idea of every algorithm performing 3D tracking is to subtract from each individual motion the average global motion of the system, in order to minimize the distance covered between two frames by the individual. The simplest case is that of standard flocks characterized approximately by straight motion: in this case one subtracts the global flock’s translation to the individual motion. This procedure is performed by all current algorithms directly on the 2D image, based on the idea that the 3D-to-2D projection of a translation is still a translation. However, this is no longer true for rotations, which change radically character when projected from 3D to 2D. A preliminary study conducted by CNR on the 2016 data shows that this phenomenon makes it very hard for the temporal linking to work on data of rotating swift flocks. More specifically, many instances of wrong linkage are produced, in which bird A at time t is linked to bird B at time t+l, giving rise to wrong trajectories