Could Your Phone Detect Your Horse's Lameness? New Research Brings Markerless Motion Capture Closer to Reality
- Human(e) Equine Project
- Aug 5, 2025
- 6 min read
Could markerless motion capture change how horse lameness is detected? A 2025 proof-of-concept study from Southwest Research Institute suggests it might. Researchers demonstrated that full-body, three-dimensional movement analysis of horses is possible using video cameras alone — no sensors, no reflective markers, no laboratory required. The system calculated joint angles across the entire horse at walk, trot, and canter with accuracy comparable to traditional motion capture methods for most joints. While the technology is not yet ready for clinical or barn-level use, it represents a meaningful step toward making detailed movement analysis accessible outside specialist facilities — with significant implications for how lameness, soundness, and performance are assessed in everyday horse care.
Why movement analysis matters for welfare
Lameness is one of the most common and consequential welfare problems in domestic horses. It is also one of the most consistently underestimated. Research has repeatedly shown that subtle gait asymmetries — the kind that precede clinical lameness and accumulate into musculoskeletal damage over time — are difficult for even experienced eyes to detect reliably. Human visual assessment of lameness is subject to expectation bias, inconsistency between observers, and genuine perceptual limitations at the threshold of subtle movement.
Quantitative gait analysis — measuring movement with technology rather than relying on observation alone — addresses these limitations directly. It can detect changes too small to see, track progression over time, and remove the subjectivity that makes early lameness so easy to miss or dismiss.
The problem is access. Current gold-standard motion capture requires specialized laboratories, trained technicians, and significant preparation time. The horse must be instrumented with over 100 reflective markers before a single stride is recorded. That kind of analysis is available to a small minority of horses — typically those in elite sport with access to university veterinary facilities. The vast majority of horses, including those whose early soundness problems would benefit most from objective assessment, never get measured at all.
This is the gap the technology in this study is designed to close.
Overview: Teaching a computer to read how a horse moves
Researchers at Southwest Research Institute adapted an existing human markerless motion capture system — previously validated for analyzing human movement in real-world environments — for use with horses. The approach works in four steps:
First, multiple synchronized video cameras record the horse moving.
Second, a neural network — a type of machine learning system trained on thousands of images of horses — identifies the location of 54 anatomical landmarks in each camera frame.
Third, the system triangulates those 2D landmark predictions from multiple camera angles into a single 3D position for each point.
Fourth, those 3D positions are fed into a digital musculoskeletal model of a horse, which calculates joint angles across the entire body for every frame of video.
The result is a complete picture of how the horse's joints are moving — shoulder, elbow, carpus, fetlock, hip, stifle, hock, and more — without anything being placed on the horse at all.
Two neural networks were tested, trained on different datasets. The second, larger dataset — which included images of over 500 horses in a wider variety of environments — produced meaningfully better results than the first, confirming that the more varied the training data, the more accurately the system can read an unfamiliar horse.
The full pipeline — from one minute of ten-camera video footage to complete kinematic data — ran in under ten minutes. The traditional alternative required two to three hours just to instrument the horse, before processing began.
Key Findings: What markerless motion capture reveals about horse lameness detection
For most joints, the markerless system performed well. Joint angles calculated from the neural network predictions matched those from traditional marker-based motion capture within 10 degrees for 25 to 32 of 35 measured degrees of freedom, depending on the gait and the neural network used. Curve similarity ratings — how closely the predicted movement patterns matched the ground truth across full strides — were good to excellent for the majority of joints.
The system performed best at the core of the horse — the back, spine, and upper limbs — and less accurately at the distal limbs, particularly the fetlocks. This is partly a geometry problem: small movements at the hoof produce large angular changes at the fetlock, and the system does not yet have sufficient precision at that level for clinical reliability.
The poll — the top of the horse's head — was consistently one of the weaker prediction points, likely because the training data did not include enough labeled images of the head from multiple angles.
There are important limitations to be transparent about. The study validated the system on a single horse. Neural network accuracy depends heavily on the quality and diversity of the images it was trained on. The system currently requires ten calibrated cameras, which is not a barn-level setup. And while the joint angle accuracy is promising for research and broad movement assessment, the margin of error in some joints — particularly the fetlocks and sacrum — may still be too large for detecting subtle lameness in clinical settings.
The researchers acknowledge all of this directly. This is a proof of concept, not a finished product. The value of the study is in demonstrating that full 3D markerless motion capture is feasible in horses at all — which was not previously established — and in laying out what needs to improve for it to become practically useful.
What this means for the future of horse welfare assessment
For riders, trainers, and owners, the direct takeaway is not "use this now" — the technology is not there yet. The takeaway is tha
t the direction of travel in equine biomechanics research is toward accessibility, and that has real welfare implications worth understanding.
Early lameness detection could become democratized. The fundamental promise of this technology is that detailed movement analysis — currently available only in specialist settings — could eventually be performed in any arena with cameras. A trainer noticing a subtle change in how a horse is moving could get objective data rather than relying on feel alone. A horse that consistently scores well in visual assessment but shows asymmetry patterns consistent with early pathology could be caught before the damage accumulates.
Objective baselines become possible at scale. One of the challenges in managing soundness is that we rarely have a detailed record of how a horse moved when healthy. If movement analysis becomes cheap and easy enough to perform routinely — before a problem develops — veterinarians and trainers would have something to compare against when something changes. That kind of baseline data is currently almost nonexistent outside research settings.
Rider influence could be quantified more easily. The study notes explicitly that the system could be used to analyze horse-rider interactions — how the rider's movement affects the horse's joint angles and movement patterns. This connects directly to the biomechanics research we have covered in earlier posts on rider seat and rein tension. Quantifying that interaction in a real training environment, without a laboratory, would be a significant research and practical tool.
Fit-to-compete assessments could become more objective. The researchers note potential applications in competition settings for pre-event soundness screening. At present, fit-to-compete evaluations are largely subjective. Movement analysis could provide an objective complement to visual assessment — not replacing veterinary judgment, but supporting it with data.
For now, the most useful thing riders and trainers can take from this research is an orientation toward existing movement assessment tools. Inertial measurement unit (IMU) systems — wearable sensors that attach to the horse — are already commercially available and provide objective gait symmetry data accessible outside laboratory settings. They do not provide the full-body joint angle detail of this markerless system, but they are a practical current option for objective lameness screening that most horse owners have never used.
The long arc of this research points toward a future where what is currently available only to elite sport horses — detailed, objective, repeatable movement analysis — becomes a routine part of how every horse is cared for. That future is not here yet. But the proof-of-concept is.
Read the original study
Shaffer SK, Medjaouri O, Swenson B, Eliason T, Nicolella DP. A Markerless Approach for Full-Body Biomechanics of Horses. Animals. 2025; 15(15):2281. https://doi.org/10.3390/ani15152281
Comments