In the ever-evolving quest to understand how animals consume and expend energy, movement remains a central focus—yet accurately quantifying this crucial biological currency has long presented formidable challenges. Scientists have traditionally relied on wearable technology, primarily accelerometers, to gauge how much energy an animal uses by tracking its dynamic body acceleration (DBA). However, this methodology fundamentally excludes approximately half of the world’s vertebrate species due to the constraints imposed by the size and weight of these physical devices. A groundbreaking study published in the Journal of Experimental Biology now revolutionizes this field by introducing a video-based, non-invasive technique leveraging 3D tracking and deep learning, opening new horizons in animal energy expenditure research.
The biology of energy dynamics is deeply entwined with animal behavior and evolution. Energy scarcity drives survival strategies, ecological interactions, and evolutionary adaptations, making precise measurement of energy usage indispensable for biology. Up to now, movement-related energy expenditure has been quantified by measuring oxygen consumption—a reliable proxy for metabolic expenditure during aerobic respiration. Methods centered on DBA correlate oxygen uptake with body acceleration, providing valuable insight into the energetic costs of distinct behaviors. Despite their proven accuracy, accelerometer-based methods face severe limitations due to the physical burden sensors impose, particularly on small animals.
The critical limitation of accelerometer-based DBA methods lies in the requirement for trackers to weigh less than 10% of an animal’s body weight to avoid impairing natural behavior. This threshold effectively excludes animals under around 100 grams, including many fish, amphibians, and small vertebrates. The physical encumbrance of accelerometers can alter behavior and locomotion efficiency, especially in aquatic environments where factors like drag play significant roles. Such constraints have left a vast swath of the animal kingdom—species that are small, delicate, or sensitive to external loads—understudied in terms of real-world energy budgets.
Addressing these challenges, researchers from the Marine Biophysics Unit at Okinawa Institute of Science and Technology (OIST), in collaboration with Professor Amatzia Genin from Hebrew University of Jerusalem, have pioneered an elegant, scalable solution: utilizing synchronized multi-angle video recordings to reconstruct 3D movement of animals and applying deep learning algorithms to estimate dynamic body acceleration. This method dispenses with the need for physical sensors, relying instead on computational tracking of key anatomical landmarks, such as the eyes and body segments, to reconstruct fine-scale movement data critical for estimating metabolic costs.
The experimental setup is surprisingly straightforward yet technologically sophisticated. By filming coral reef damselfish (Chromis viridis) in controlled aquatic environments with multiple cameras, the team captured precise footage from different angles. They then employed deep neural networks—trained on annotated video frames—to identify and track body features. These positional data were translated into acceleration metrics that closely replicate the accelerometer-based DBA traditionally used. Validation experiments demonstrated strong correlations between video-based DBA and oxygen consumption, confirming the accuracy of the approach.
This breakthrough transcends traditional limitations, enabling energy expenditure studies on smaller species previously unreachable by accelerometer-based tracking. It also eliminates the physical interference of equipment, thus preserving authentic behavior and locomotion. By leveraging computer vision and machine learning, this method scales effortlessly for use in both laboratory and field settings, promising expansive applicability in behavioral ecology, physiological studies, and evolutionary research.
One of the most tantalizing applications highlighted is the potential to unravel energetic dynamics in collective behavior, particularly schooling in fish. Historically, the energy costs to individuals within schools were enigmatic due to measuring challenges in natural environments. The video-based DBA technique allows researchers to distinguish energetic contributions from leading versus trailing fish and assess whether schooling confers substantial energy savings. Such insights could shed light on the evolutionary drivers and ecological functions of this widespread, complex behavior.
In ecological and evolutionary perspectives, energy efficiency underpins fitness and survival. With finer-scale energy expenditure measurements now accessible across a broader spectrum of wildlife, scientists can more deeply probe how different species optimize their behavior, adapt to environments, and evolve metabolic strategies. This tool opens avenues to assess, for example, how fish manage energy under varying flow speeds or predator pressures, linking physiology with ecosystem dynamics across spatial and temporal scales.
Moreover, the deployment of advanced machine learning within behavioral monitoring signals a larger trend in biological research—combining big data, AI, and traditional physiology to transcend previous technical constraints. By embracing these interdisciplinary approaches, biology stands on the cusp of transformative discoveries about life’s fundamental energetic processes, potentially translating into conservation strategies by detailing species-specific energy budgets under environmental stressors.
This innovative technique also alleviates ethical and logistical concerns inherent in animal experimentation. Without attaching physical devices, animals are less stressed, and experimental conditions more naturalistic. Remote monitoring via video reduces human interference, enabling continuous, unobtrusive observation in both captive and wild contexts. Hence, this approach aligns with evolving standards for animal welfare and experimental fidelity.
As deep learning algorithms continue to mature, the precision and applicability of video-based DBA estimates will only improve. Future enhancements could allow identification and quantification of subtler movements, integration with environmental sensors, or adaptation to other taxa beyond fish, including terrestrial vertebrates or invertebrates. This opens a rich landscape for research that bridges physiology, behavior, ecology, and technology.
In sum, the video-based dynamic body acceleration methodology represents a paradigm shift in the study of bioenergetics. It leverages accessible video capture and state-of-the-art machine learning to quantify energy use with unprecedented inclusiveness and precision. This work propels the biological sciences towards a more comprehensive understanding of how movement shapes energy demands across the diversity of life, ultimately enriching our understanding of animal ecology and evolution.
Subject of Research: Animals
Article Title: Use of videos to measure dynamic body acceleration as a proxy for metabolic costs in coral reef damselfish (Chromis viridis)
News Publication Date: 10-Apr-2025
Web References: http://dx.doi.org/10.1242/jeb.249717
Image Credits: Ishikawa et al., 2025
Keywords: Animal locomotion, Ecological methods, Evolutionary methods, Deep learning, Evolutionary processes, Fish, Animal experimentation, Field experiments, Ecological processes, In vivo imaging, Metabolism, Physiology
