
When body parts occlude each other or overlap with cage walls and equipment edges, keypoint and contour detection is easily lost, reducing analysis stability.
China's first markerless 3D animal pose capture system. Identify 16+ keypoints, quantify 40+ behavior subtypes, and export publication-ready charts — for mice, rats, NHPs, and more.

China's first 3D-AI fine ethology product — backed by publications, patents, media coverage, and peer recognition






6 granted · behavior recognition · 3D capture · quantification
In a recent wild monkey behavior study, our work received an extensive dedicated commentary:
Read Original
“Adopting automated behavioral recognition is scalable, increasing the speed, quantity, and detail of data that can be collected and analyzed. Once classifiers have been trained, such work can move beyond broad classification of general behavioral states (eating, resting, etc.) to include fine-grained analysis at multiple layers/dimensions of behavior.”
Related work appears in Nature family journals, with coverage from CAS, China Science Daily, CCTV, and dozens of outlets — featured as a CAS progress highlight and reviewed by China Brain Project experts.
Combining markerless 3D capture, AI recognition, and multi-species support to unlock the full potential of 3D behavior analysis.
Severe information loss — fails to capture real behavior

When body parts occlude each other or overlap with cage walls and equipment edges, keypoint and contour detection is easily lost, reducing analysis stability.
Tracking relies only on centroids or outlines — limited behavioral dimensions, unable to capture posture changes, limb motion, or complex behavior structure.

Many behaviors look similar in top-down view — immobility, sniffing, grooming, mild exploration — and are easily misclassified or merged into one category.

Outputs only simple metrics such as distance, speed, dwell time, and zone entries — insufficient to capture complex behavioral changes.
Fine-grained, comprehensive, objective behavior characterization

Multi-camera synchronized capture from different angles reduces single-view occlusion and information loss for more complete locomotion data.

Reconstructs 3D skeletons from multi-view video — accurately restoring spatial position and posture of head, trunk, limbs, and tail.

Powered by BehaviorAtlas decomposition — automatically segments continuous video into behavior episodes and types for fine-grained spectrum analysis.

Beyond speed, distance, and time — quantifies posture, joint angles, body height, trajectories, behavior proportions, temporal dynamics, and more.
Multi-species solutions from rodents to primates for comprehensive behavior research.


Fine behavior analysis for mice and rats — supporting disease models, drug screening, and mechanism studies.
3D Skeleton Reconstruction:4-camera sync capture with multi-view 3D reconstruction — mm-level accuracy with occlusion handling.
Advanced Deep Learning:Identifies 16 keypoints across body, limbs, head, and tail — deep learning trained on large-scale experimental data for stable pose estimation.
Rich Application Support:Supports disease model characterization, behavior modulation, and efficacy evaluation.
Explore full capabilities, workflows, and multi-species solutions in the 3D-AI Fine Behavior Software Suite
Real-world application data and research outcomes of BehaviorAtlas 3D-AI across global scientific settings.
10 papers pending publication







Real-world research cases powered by BehaviorAtlas 3D-AI — spanning autism, anxiety, depression, Parkinson's, Alzheimer's, and acute stress models.
Selected publications powered by our 3D-AI system — supporting teams in publishing 50+ high-impact papers in Nature Aging, STTT, and other top journals.