Quantifying Insect Behavior Across Scales Using Computer Vision

Author: Sharma, Tarun

Year: 2026

Degree: Dissertation (Ph.D.)

Advisors: Parker, Joseph; Dickinson, Michael H.

Committee Members: Siapas, Athanassios G.; Parker, Joseph; Dickinson, Michael H.; Hong, Elizabeth J.

Option: Computation and Neural Systems

DOI: 10.7907/yphx-fh43

Abstract

Insects exhibit extraordinary behavioral diversity from rapid sensorimotor control required for flight to coordinated behavior across large societies. While some behaviors are best studied in controlled lab settings and require precise fine-grained measurements, others demand long-term observations in more natural settings that cannot be replicated in lab. In this thesis, I demonstrate the applications of computer vision to quantify insect behavior across scales, from the level of individuals to colonies, in both lab and field conditions.

In the first part, I examine the role of mechanosensory cells, campaniform sensilla, on the stabilization response of the fruit fly, Drosophila melanogaster. Using tethered flies mounted on a rotating arena, I measure head and wing equilibrium responses using marker less pose estimation and edge tracking. By genetically silencing different subsets of campaniform sensilla, I show a linear relationship between the number of cells silenced and the magnitude of head and wing response, providing direct experimental evidence for their role in the equilibrium response.

In the second part, I apply computer vision techniques to study colony scale movement dynamics at the nest entrance of the ant species, Liometopum occidentale, in the Angeles National Forest. I introduce a custom, low-cost, field deployable multisensory camera trap, called the Ethocam, and collect an extensive dataset of hourly videos from three ant nests spanning over 100 days. Using computer vision methods for detection and multi-object tracking, I quantify circadian activity patterns, directional traffic imbalances, environmental drivers of activity and walking speed, and spatiotemporal movement dynamics across distinct trails.

Together, by integrating lab experiments, long term field data, and quantitative analysis using computer vision, this thesis presents a general framework for studying insect behavior across scales in both controlled and natural environments.

Files