A Review of the Evolution of Rodent Tracking and Behaviour Analysis

Advances in computational power, miniaturisation, and cost reduction of electronic devices have profoundly transformed behavioural neuroscience, enabling automated, high-throughput data acquisition. In rodent behaviour analysis, computer vision and sensor-based systems now allow continuous tracking and pose estimation over extended periods, reducing human bias and improving reproducibility. Over the past 25 years, tracking methods have evolved from simple 2D single-animal approaches to sophisticated multi-animal systems capable of markerless identification and 3D pose estimation. These developments were driven by innovations in image processing, physics-based modelling, and, more recently, deep learning architectures that enable precise skeletal estimation and real-time inference. Behaviour recognition has similarly progressed, moving beyond rule-based systems to supervised machine learning and unsupervised clustering techniques that uncover latent behavioural patterns without predefined categories. This review retraces these technological and algorithmic milestones, highlighting how they shaped modern computational ethology and opened new avenues for studying complex social interactions and disease models. By examining trends and challenges, we provide insights into future directions for scalable, interpretable, and consensus-driven behaviour analysis.