Hummingbirds hover, enabling them to serve as exclusive pollinators for many plant groups with which they have coevolved. They combine bird muscle power with insect flight skills to hover and maneuver in any direction. This study investigates an under-explored aspect of their hovering flight: many species exhibit an alternative gait, in which they flap their wings discontinuously, including momentary pauses sporadically interspersed between short series of wingbeats, which collectively generate a pattern we denote as 'intermittent hovering'. These brief pauses have not been characterized in hummingbirds but may play aerodynamic, energetic, and/or signaling roles. To detect and measure intermittent hovering, we present a high-throughput method to quantify these irregular wingbeat pauses using computer vision and signal analysis. High-speed videos of 11 species collected by recording free-living hummingbirds allowed us to track and analyze the pauses during hovering. The proposed algorithm has a precision of 74%, accurately detecting flapping pauses and thus intermittent hovering. Most of the misclassification errors were false positives: when very still hovering hummingbirds were continuously moving their wings, but some of the consecutive frames were similar enough that the algorithm classified them as a brief pause. These false positives, however, are easily discarded upon quick visual inspection, and the algorithm had only 19 false negatives across all videos, actually detecting intermittent hovering if it was present. This method is a step forward in creating tools that help researchers analyze complex behavioral patterns. Our study confirms the feasibility of reliably detecting intermittent hovering, contributing to our understanding of hummingbird flight dynamics.
Keywords: Computer vision; Deep learning; Hummingbirds; Pausing behavior; Zero learning.
© 2026. Published by The Company of Biologists.