Tracking the Flight: Exploring a Computational Framework for Analyzing Escape Responses in Plains Zebra (Equus quagga)

TL;DR

This paper tackles how to quantify escape dynamics in a natural zebra herd using drone footage by evaluating three trajectory-unwrapping pipelines: frame-to-frame image registration, interpolated Structure-from-Motion (SfM), and a hybrid approach. Using a field dataset of 44 zebras in a four-wave escape from Mpala, Kenya, the SfM+Linear Interpolation method yielded the most accurate world-coordinate trajectories, with a mean dispersion of $0.275$ body lengths compared to $0.910$ for frame registration. The analysis reveals high herd polarization during rapid movement, a brief spacing expansion before stopping, and stronger alignment among centrally located individuals, illustrating scalable metrics for collective behavior. The framework, implemented in open-source tools, demonstrates potential to scale to larger datasets and cross-taxa studies, enabling broader investigations of escape strategies and social coordination in wild animals.

Abstract

Ethological research increasingly benefits from the growing affordability and accessibility of drones, which enable the capture of high-resolution footage of animal movement at fine spatial and temporal scales. However, analyzing such footage presents the technical challenge of separating animal movement from drone motion. While non-trivial, computer vision techniques such as image registration and Structure-from-Motion (SfM) offer practical solutions. For conservationists, open-source tools that are user-friendly, require minimal setup, and deliver timely results are especially valuable for efficient data interpretation. This study evaluates three approaches: a bioimaging-based registration technique, an SfM pipeline, and a hybrid interpolation method. We apply these to a recorded escape event involving 44 plains zebras, captured in a single drone video. Using the best-performing method, we extract individual trajectories and identify key behavioral patterns: increased alignment (polarization) during escape, a brief widening of spacing just before stopping, and tighter coordination near the group's center. These insights highlight the method's effectiveness and its potential to scale to larger datasets, contributing to broader investigations of collective animal behavior.

Figures (4)

• Figure 1: Stitched drone flight path overlaid with the movement trajectories of 44 individual zebra within the observed herd. The trajectories are computed using the SfM linearly-interpolated approach.
• Figure 2: Scatter plot showing co-evolution of average group speed (body lengths/s) and alignment (polarization) in in $\sim$1-second intervals (30 frames), colored by time. Arrows trace temporal progression, highlighting shifts between fast, aligned and slow, unaligned movement. Smoothed with a Savitzky-Golay filter (window size 7); key regions annotated.
• Figure 3: Plot showing individual zebra alignment with the herd’s mean direction over time. Each row is a zebra; color indicates the cosine of the angular difference between its orientation and the group’s (red: opposite, white: perpendicular, blue: aligned, grey: missing).Values range from -1 (opposite) to 1 (aligned).
• Figure 4: Frame-by-frame Pearson correlation between each zebra’s distance from the herd centroid and its alignment with the average herd direction. Each point shows whether individuals farther from the center are more or less aligned in that frame.