Movyent: A Browser-Based Platform for Accessible CPU-Based Animal Behavioral Tracking

Quantitative behavioral analysis in neuroscience has traditionally relied on expensive commercial software or complex open-source tools requiring advanced programming knowledge and GPU acceleration. To address these accessibility barriers, we developed Movyent, a free browser-based platform for automated animal behavioral tracking that operates entirely on a local CPU. The core computer vision engine performs contour-based tracking using the Mixture of Gaussians 2 (MOG2) background subtraction algorithm and optionally provides keypoint-based posture estimation using the YOLOv8-Nano model. The system calculates standard behavioral indicators such as velocity, thigmotaxis, and freezing, and ensures data integrity by maintaining missing values when occlusion occurs. Complex behaviors such as rearing and grooming are identified through explainable rule-based classifiers with explicitly defined kinematic thresholds. Movyent was evaluated using 50 standard 1080p Open Field Test videos and achieved a 98.5% detection success rate, Pearson correlation coefficients of 0.974 for total travel distance and 0.952 for instantaneous velocity compared to manual scoring, and 92% accuracy for freezing behavior detection. On an Intel i5-class laptop, the system processed video at 21 frames per second with memory usage below 500 MB. These results demonstrate that reliable automated behavioral analysis is achievable in resource-limited computing environments, providing a transparent and accessible alternative to costly commercial platforms and technically demanding deep learning tools.