Computer vision carve analysis
Read
Your line
Upload one carve. Board Vision tracks 33 body landmarks through every frame and measures whether your legs drove the turn and your torso followed — the timing that separates a carve from a muscled turn.
carve clip · frontside
One carve per clip · frontside or backside · on the web now, iOS soon
What it measures
Three signals,
every frame
Thirty-three landmarks per frame reduce to three measurements in true image pixels. Each one is smoothed, then read across the whole carve.
Compression
Ratio · 1.0 extended → 0.5 deep
How deep you load the support leg, tracked frame by frame as a ratio of leg extension — straight hip-to-ankle distance over the bent path through the knee.
Twist
Normalized 0 → 1
How far your shoulders wind off your hips, with a foreshortening fallback for when you're facing the camera and the angle between the two axles stops being trustworthy.
Lean
Signed degrees off vertical
How far you lay the body over from your knees, not your hips — so a hinge at the waist doesn't get mistaken for bank. The sign is camera-relative.
How it works
Four steps,
one carve
Film one turn
Whole body in frame, start to finish. Shoot it on whatever you've got.
Tag and upload
Frontside or backside, then send the clip. Rotation is baked into the pixels and the clip is resampled to a constant 24 fps.
Pose tracking reads it
33 landmarks per frame, lined up 1:1 with the pixels — and your clip comes back with the skeleton overlaid, frame-accurate.
The two curves get compared
Board Vision cross-correlates leg extension against torso rotation over ±5 frames and tells you which led, and how well they tracked together.
The read-out
Deterministic,
not generated
Two numbers come out of the correlation — which movement led, and how tightly the curves tracked. They land on one of six read-outs. Geometry and correlation, not a language model writing encouragement.
Legs led · Connected
Compression /
rotation correlation
0.87
Compression /
rotation lag
3 fr
Analyzable
coverage
94%
Frames
analyzed
118/126
Six possible read-outs
Lag decides the column. The coefficient decides the row.
| Legsled | Synch-ronized | Torsoled | |
|---|---|---|---|
| Connected | This clip | Not this clip | Not this clip |
| Disconnected | Not this clip | Not this clip | Not this clip |
This clip: legs led by 3 frames, connected.
Per-frame signals · shared time axis
Three different units — never one shared scale.
Frame 063126
Honest limits
What it
doesn’t do
A narrow measurement you can trust beats a broad one you can’t.
One carve per clip
Frontside or backside. Carves only for now — snaps, wraps and cutbacks are coming.
There is no score
It measures mechanics and timing, not style points. No grade, no rating, and nothing measured against anyone else's turn.
It tells you when it can't see
Every read-out ships with its coverage, per signal, so you know which body part left the frame. Under 60% it turns the clip down rather than guessing.
Pricing
Same analysis
either way
Pro is about volume, not features. Every clip gets the same three signals, the same read-out and the same skeleton overlay.
Free
$0
NO CARD, NO RENEWAL
- Three analyses, one carve each
- All three signals, full read-out
- Frame-accurate skeleton overlay
- Coverage report on every clip
Pro
For every session$9.99/ MONTH
BILLED MONTHLY, CANCEL ANY TIME
- Thirty analyses a month
- Everything in Free — same analysis, same depth
- Allowance resets monthly
- Cancel whenever — refused clips never count
Prices shown in your local currency, tax included.
FAQ
Straight
answers
- What kind of clip works best?
- One carve, whole body in frame from start to finish, shot side-on if you can. Board Vision runs in the browser today, with iOS to follow once the App Store approves it. Either way the clip is normalized before anything is measured: rotation baked into the pixels, a constant 24 fps, long side capped at 1280.
- Why did it refuse my clip?
- Coverage came in under 60% — too few frames where every signal was readable. Usually a limb out of frame, a rider too small in the shot, or spray over the body. Board Vision would rather turn the clip down than guess at the gaps.
- Does it compare me to pro surfers?
- No. There's no baseline and no grade. It reads your carve on its own terms: which movement led, by how many frames, and how tightly the two curves tracked as one motion.
- What is coverage?
- The share of frames the tracker could actually read — every signal at once, plus how often it saw a body at all, plus a figure per signal so you know which body part to keep in frame next time. It's evidence, not a grade: a high-coverage reading of a badly ridden carve is normal and correct.
- Which maneuvers are supported?
- Carves, one per clip, tagged frontside or backside. Snaps, wraps and cutbacks are coming.