Tracking Spatial and Activity Patterns in Captive Reptiles Using Deep Learning

The knowledge base for many small vertebrate species remains limited, largely because traditional manual data collection methods often overlook less charismatic species, such as reptiles. To address this, our pilot study harnesses open-source deep learning and markerless pose estimation technologies to evaluate the technical feasibility of tracking the spatial use and activity profiles of captive ectotherms. Specifically, we tracked these patterns over two months in a dynamically modified environment for Australian barking geckos (Underwoodisaurus milii). Our findings reveal descriptive changes in spatial occupancy and proximity across varying structural layouts. The system achieved a high raw detection accuracy (96.4%) and spatial categorization accuracy (91.7%) when validated against manual ground-truth data, confirming its robust technical performance and precision. Additionally, we automatically evaluated spatial proxies such as activity time budget, velocity, acceleration, and height usage, standardizing the analysis of extensive video recordings for nocturnal species. This pilot test introduces a simple, cost-effective method for rapid data extraction, offering a reliable, scalable monitoring solution for the management of understudied species.