EVERY
MOVEMENT
LEAVES A
SIGNAL.
One future movement-intelligence layer for athletes across sports. We're exploring what ordinary phone footage can reveal through pose, geometry, and temporal signals.
NOT IDENTITY13
KEYPOINTS*
SEE THE
SIGNAL.
Switch visual layers to inspect the concept. The interactive graphic is an illustration, not measured tracking from the research clip.
Sport modes change the illustrative phase label; they do not run a sport-specific model.
In the supplied MediaPipe CPU results, the selected person's right-elbow angle was 144.94° at frame 194.
Recorded measurement from the supplied run; the illustration on the left is not derived from that frame.Watch the court footage used in NextPlay's research. The visual study above is illustrative and is not generated from this video.
FOOTAGE IS JUST
THE BEGINNING.
FOOTAGESUPPLIED COURT VIDEO
FRAMESMEDIAPIPE SAMPLE RUN
ANGLESMEASURED FRAME SIGNALS
RECORDPLANNED INTEGRATION
MORE THAN
A SILHOUETTE.
From the supplied CPU run: 13 keypoints, eight joint angles, and per-angle minimum, maximum, mean, and range over 209 frames. These are geometric signals, not validated coaching judgments.
UPPER BODY
Shoulder, elbow and wrist chains reveal frame-level geometry.
↗HIPS + KNEES
Selected landmark relationships and angle changes across samples.
↗TEMPORAL RANGE
Min, max, mean and range preserve how signals changed across time.
↗ONE VIDEO.
527.8
SECONDS.
Keypoint R-CNN inference was the primary cost in a measured, CPU-only benchmark using 288 sampled frames.
*Both engines processed the same 288-frame sample on CPU at FP32, batch size 1. Different model architectures; this is a processing-time comparison, not accuracy or coaching-quality validation.
PROVE FIRST.
SCALE NEXT.
We are building on measured foundations. GPU acceleration, richer models, and athlete-record integration remain planned work.
Deterministic frame selection · MediaPipe CPU baseline · Keypoint R-CNN comparison · output checks · reproducible CPU workflow.
✓NVIDIA GPU profiling · batch/precision testing · athlete record integration · useful movement summaries · pilot evaluation.
↗Sport-specific models · longitudinal comparison · production inference · concurrent workloads · edge evaluation.
○WHAT WE CANNOT CLAIM YET.
BRING A
REAL
TRAINING
QUESTION.
We welcome careful R&D conversations with athletes, coaches, programmes, technology partners, and research teams.
bhagyashrimeena@nextplaysports.in ↗