Send a short video of any pet in motion. Get back a skeleton overlay, frame-level keypoints and 90+ biomechanics fields across 13 metric families. We give you the hard computer-vision layer, animal-tuned and production-ready, so you build the product.
One call, five outputs
The original clip with the tracked skeleton drawn on top. Drop it straight into your UI or a report.
Frame-level keypoint coordinates and confidence for every tracked joint. The raw signal, if you want to build your own metrics.
90+ derived fields across 13 families: stance, stride, joint angles, head nod, hip hike, velocity, weight-bearing and more.
Per-cycle start / end, duration and stride length, with short- and long-stride cycles flagged automatically.
A normalized, plain-language summary you can render in an app, dashboard, clinical report or partner system.
Structured output
Metrics come back flat-namespaced as family.field. Spatial values are body-length normalized, so results compare cleanly across breeds and body sizes. Below is a trimmed slice of a real response.
On a clean clip the engine can emit well over 90 raw biomechanical fields. Exactly which fields populate depends on video angle, clarity, occlusion, walking speed and keypoint confidence.
{
"scale": {
"body_length_px": 412.0,
"motion_axis_stability": 0.94,
"shoulder_height_bl": 0.61
},
"quality": {
"mean_keypoint_confidence": 0.88,
"usable_frame_ratio": 0.91
},
"stance": {
"front_left_stance_fraction": 0.62,
"front_right_stance_fraction": 0.55,
"front_stance_fraction_asymmetry": 0.12
},
"stride": {
"cadence_steps_per_min": 108.4,
"mean_stride_length_bl": 1.18,
"front_stride_length_asymmetry": 0.09
},
"head_nod": {
"head_nod_asymmetry_bl": 0.041,
"head_nod_side": "right_front_painful"
},
"weight_bearing": {
"front_left_weight_index": 0.54,
"front_right_weight_index": 0.46,
"front_weight_index_asymmetry": 0.08
}
}13 families · 90+ fields
The biomechanics layer is what you would otherwise spend quarters building. It is ours, and it ships with every call.
13
metric families
90+
biomechanics fields
4
limbs tracked
scaleBody-length pixel scale, motion-axis angle and stability, shoulder height.
qualityKeypoint confidence, low-confidence ratio, completeness, usable-frame ratio.
stancePer-limb stance / swing fraction and time, plus left/right asymmetry.
stridePer-limb stride time, length and consistency, cadence, stride asymmetry.
stride_timelineEvery detected stride cycle with timing, length and flagged anomalies.
joint_anglesPer-limb knee angle mean / min / max, range of motion and asymmetry.
head_nodHead-nod asymmetry and the inferred painful front-limb side.
hip_hikeHip-hike asymmetry and the inferred painful hind-limb side.
velocityMean / max velocity, variability and steady-state fraction.
back_swayLateral sway and vertical oscillation of the back line.
stepFront and back step separation and step width.
symmetryFront and back paw-lift range asymmetry.
weight_bearingPer-limb weight-bearing proxy index and front / back asymmetry.
Build beyond gait
The same pose and biomechanics primitives power very different products.
Surface limping, asymmetry and reduced range of motion before they are obvious to the eye.
Track posture, movement and activity in cats and dogs for research or product features beyond gait.
Compare the same animal across sessions to quantify recovery and treatment response.
Generate structured, normalized kinematic data at scale from ordinary phone video.
Attach objective, repeatable mobility evidence to claims and wellness programs.
Power decision-support tooling for clinics with consistent, explainable metrics.
Why build on Peqaboo
Built for cats, dogs and small animals, not retrofitted from human pose. It handles four-legged movement, tails and varied body shapes.
Confidence scoring, usable-frame detection and auto-rejection of unusable footage. You never ship a metric computed from garbage frames.
The 13 metric families are the real value: keypoints turned into stance, stride, ROM, head nod, hip hike and weight-bearing. You do not rebuild this.
Spatial metrics are normalized to body length, so a Chihuahua and a Great Dane are directly comparable.
Beyond raw numbers: inferred painful side and left/right asymmetry indices, ready to act on.
Optionally ground reports in Peqaboo’s veterinary and behavioural knowledge base for richer, safer output.
Runs on managed H100 GPUs. You send a URL or a file and get results back. No models to host, no GPUs to babysit.
Overlay video, JSON and report summaries out of the box, so you can ship a product in days, not quarters.
Pricing
Usage is metered in credits. One credit covers five seconds of analysed video, so a 10-second clip is two credits. You are only charged when a result is delivered.
Pay as you go
Per credit
Top up credits, no commitment. Standard output tier, scales with you.
Request API accessRequest a sandbox key, run your own clips, and see the full output before you commit.