AI Swimming Video Analysis. Above and Below the Surface

For Swim Tech Startups and Elite Academies. We build custom Computer Vision Engines that track stroke mechanics, turn times, and velocity in real-time—solving the complex challenges of water refraction and splash occlusion.

Why Water Breaks Standard AI.

Tracking a human in a pool is one of the hardest challenges in Computer Vision. Off-the-shelf pose estimation models (like standard MediaPipe) fail immediately due to three physics blockers:

The Refraction Barrier (1)

The Refraction Barrier

Light bends when it hits water. A standard model thinks the swimmer's arm is 6 inches away from where it actually is. We build Refraction-Correction Algorithms to map true anatomical position.

Occlusion by Splash (1)

Occlusion by Splash

Bubbles and whitewash hide the swimmer's body. Our Temporal Logic Models predict limb position based on stroke cadence, filling in the gaps when the camera is blinded.

Dual-Medium Tracking (1)

Dual-Medium Tracking

Tracking a swimmer as they break the surface (air-to-water transition) usually crashes the model. Our engines fuse above-water and underwater feeds into a single continuous biomechanical profile.

Lack of Reliable Training Data

Lack of Reliable Training Data

Swimming datasets are limited and difficult to capture due to underwater conditions, motion blur, and inconsistent labeling. This makes it challenging to train accurate, production-ready AI models.

Solutions for the Swimming Ecosystem

From "Smart Pool" apps to Olympic-grade performance lab

Mobile Stroke Coach (B2C)

Mobile Stroke Coach (B2C)

For consumer apps. Users place their phone at the end of the lane. Our AI counts laps, calculates SWOLF scores, and analyzes head position without requiring a smart watch or wearable.

https://aiml-staging.folio3.site/swimming-software

Multi-Cam Performance Labs

For elite academies. We stitch feeds from underwater and overhead cameras to generate a 3D Digital Twin of the swimmer, analyzing drag coefficients and streamline efficiency.

https://aiml-staging.folio3.site/swimming-software

Automated Drowning Detection (Safety)

For smart pools. Beyond performance, we build Safety Vision Layers that detect motionless swimmers on the pool floor and trigger alerts to lifeguards in <3 seconds.

Turning turbulent water into structured data.

How We Engineer Through the Noise

Step 1_ Environment Calibration (1)

Step 1: Environment Calibration

We calibrate the model for specific pool conditions (lighting, water clarity). We use "Background Subtraction" to isolate the swimmer from the lane ropes and tiles.

Step 2_ Refraction-Aware Pose Estimation (1).

Step 2: Refraction-Aware Pose Estimation

We train custom models on underwater datasets. This ensures the AI recognizes a "distorted" arm as a human limb, correcting the coordinates to match physical reality.

Step 3_ Cycle & Phase Detection (1)

Step 3: Cycle & Phase Detection

Our Logic Engine identifies the four phases of a stroke (Catch, Pull, Exit, Recovery). This allows us to calculate DPS (Distance Per Stroke) and specific phase velocity.

Step 4_ Edge Deployment (1)

Step 4: Edge Deployment

Pools often have poor Wi-Fi. We optimize models to run offline on tablets (iPad Pro) or edge gateways so coaches get feedback instantly on the pool deck.

Customer Story

Container (4)

An image of a "Shot Heatmap" dashboard on an iPad.

"We wanted to build a 'Smart Gym' experience without rewiring the facility. Folio3 built a vision system that runs on standard security cameras, automatically tracking stats for every player who steps on the court." — CTO, Basketball Analytics Startup.
FAQ SECTION

Frequently asked questions

For biomechanics, yes, underwater views are best. However, for lap counting, stroke rate, and speed tracking, we can achieve high accuracy using standard overhead or pool-deck smartphone cameras.
Yes. Our "Re-Identification" algorithms track individual swimmers even when they cross paths or circle swim, maintaining separate data profiles for each athlete.
Yes. We build custom IP. You own the model, the training data, and the application code. You are not renting a SaaS product.
Yes. We are a full-stack agency. We can build the vision engine and the user-facing iOS/Android app (React Native/Flutter) for swimmers and coaches.
Our models achieve frame-level accuracy within 2–3% of expert manual annotation on benchmark datasets, and improve further when fine-tuned on your specific pool environment.
Yes. We can build data bridges to ingest split timing data and merge it with our vision-based biomechanics output into a unified athlete profile.

Ready to Digitize the Pool?

Whether you are building a safety tool or a performance app, we have the engineering team to execute.

Contact

Let's get in touch

Fill the form below or Contact us at +1 408 365-4638 / email us via contact@folio3.ai

This site is protected by reCAPTCHA and the Google
  • 22+ Years

    of Experience In the AI Domain

  • 950+ Projects

    Delivered Worldwide

  • 99%

    Client Satisfaction

  • Est. 1995

    Founded

  • Same Day

    Response Guaranteed

Support

Contact Info

+1 408 365-4638
contact@folio3.ai

Map

Visit our office

6701 Koll Center Parkway, #250 Pleasanton, CA 94566