SKI FORM
HOME IN DEV
REAL OUTPUT · NIGHT RUN
↓ SCROLL TO EXPLORE ✕ FRAME-BY-FRAME TECHNIQUE ANALYSIS IN DEVELOPMENT ●

Computer vision
meets the
fall line.

SEE HOW
IT WORKS

From a real run.

SAMPLE — PIPELINE OUTPUT

Turn 2, a right turn, with the pose overlay
T02 · RIGHT
Turn 3, a left turn, with the pose overlay
T03 · LEFT
Turn 4, a right turn, with the pose overlay
T04 · RIGHT
Turn 5, a left turn, with the pose overlay
T05 · LEFT
REAL OUTPUT

Night Run Overlay

A 29-second night run, filmed on a 360° camera. The pipeline tracked the skier through the run, drew the pose skeleton and the center of mass, and split the run into turns. The clip above shows 16 seconds of it, and these frames are the turns in that clip.

TURNS FOUND
9
DURATION
29 s
FRAMES
871
AVG. CONFIDENCE
0.90
NOW—

It starts with footage you already shoot.

01 — CAPTURE

ILLUSTRATIVE

Capture

Upload a run filmed on a 360° action camera. The pipeline finds the skier in every frame and follows them down the hill, so nobody has to chase you with a camera.

WorkingPIPELINE
Python360° videoDetection
STAGE 01 OF 06

Every frame. Every joint.

02 — POSE

ILLUSTRATIVE

Pose Estimation

A pose model places body keypoints on every frame: head, shoulders, hips, knees, ankles. Each point carries a confidence score, and the raw signals are smoothed before anything is measured.

WorkingPIPELINE
Pose estimationKeypointsSmoothing
STAGE 02 OF 06

Finds every turn on its own.

03 — SEGMENT

ILLUSTRATIVE

Turn Segmentation

A skier's center of mass swings side to side down the fall line. The pipeline tracks that lateral path and splits the run into individual turns, with no manual tagging.

WorkingPIPELINE
Signal processingTime seriesSegmentation
STAGE 03 OF 06

One score per turn, and the reasons why.

04 — SCORE

ILLUSTRATIVE

Rubric Scoring

Each turn is graded against a weighted rubric of technique metrics: the things a coach watches for, turned into numbers. The metrics and weights are still being tuned, so they stay under wraps for now.

CalibratingPIPELINE
Feature engineeringBiomechanicsScoring
STAGE 04 OF 06

See it on the video, not in a spreadsheet.

05 — REVIEW

ILLUSTRATIVE

Overlay + Dashboard

Results come back as an annotated overlay video and a web dashboard: a library of runs, a turn-by-turn breakdown, and trends across many runs.

WorkingFULL STACK
FastAPISQLiteReactViteTailwind
STAGE 05 OF 06

Next: a sensor on the boot.

06 — HARDWARE

CONCEPT

Boot Pod

A camera sees the body, but it cannot feel the ski. A small motion-sensor pod on each boot will record what the video misses, synced to the footage. Two mount positions are under test: the cuff power strap and the lower shell.

In R&DHARDWARE
IMUBLEFirmwareSensor fusion
STAGE 06 OF 06

Ski Form is a side project that turns ski footage into turn-by-turn technique feedback. Software first, hardware next.

A FULL SITE IS ON THE WAY. UNTIL THEN, SOME DETAILS STAY INTENTIONALLY VAGUE.

01 / COMPUTER VISION

Person detection, pose estimation, keypoint confidence, video overlays

02 / SIGNAL PROCESSING

Smoothing noisy keypoint signals, time-series segmentation, feature extraction

03 / BACKEND

Python, a multi-stage analysis pipeline, FastAPI, SQLite, self-hosting

04 / FRONTEND

React, Vite, Tailwind, data-heavy dashboards and video review tools

05 / HARDWARE

IMU sensing, Bluetooth Low Energy, firmware, video and sensor time sync

06 / PRODUCT

Turning coaching knowledge into a measurable rubric and a phased roadmap