User Guide

Complete guide to using Swim ProHub 360 for swimming video analysis

Getting Started

Welcome to Swim ProHub 360

Swim ProHub 360 is an AI-powered swimming analysis tool that uses computer vision to analyze your swimming technique. Upload a video of your swim and receive detailed metrics, AI-generated insights, and personalized coaching advice.

Quick Start Guide

Record Your Swim

Use a camera or smartphone to record your swimming. For best results, record from the side of the pool with a clear view of your full body.

Upload Your Video

Drag and drop your video file onto the upload area, or click to browse and select your file.

Wait for Analysis

The AI will process your video, detecting your body position and movements frame by frame. This typically takes 1-3 minutes.

Review Your Results

View detailed metrics about your stroke technique, body position, and efficiency. Use the AI Coach to ask questions about your technique.

Video Upload

Video Requirements

📁
File Size
Max 100 MB
🎬
Formats
MP4, MOV, AVI, MKV, WebM
⏱️
Duration
Max 5 minutes
📐
Resolution
720p+ recommended

Recording Tips for Best Results

  • Camera Position: Place the camera at pool deck level, perpendicular to your swimming direction
  • Lighting: Ensure good lighting - outdoor pools in daylight work best
  • Stability: Use a tripod or stable surface to avoid shaky footage
  • Full Body: Make sure your entire body is visible in the frame
  • Distance: Keep 3-5 meters away from the swimmer for optimal detection
  • Water Clarity: Clear water helps with better pose detection
Side-view recordings typically yield the most accurate stroke analysis. Underwater footage may have reduced accuracy due to light refraction.

Understanding Your Metrics

💡 Stroke-Specific Ranges: Optimal metric ranges vary significantly by stroke type. When you analyze a video, the app automatically detects your stroke and shows you the appropriate target ranges. Below are general guidelines - your results page will show stroke-specific targets.

Stroke Mechanics

Stroke Style

The AI automatically detects which stroke you're swimming: Freestyle, Backstroke, Breaststroke, or Butterfly.

Stroke Rate

Measured in: strokes per minute (spm)
How many complete stroke cycles you perform per minute. Higher isn't always better - efficiency matters more.
Freestyle/Backstroke: 45-70 spm | Breaststroke: 30-50 spm | Butterfly: 40-60 spm

Stroke Count

Total number of strokes detected in the video. Use this to track consistency across similar distances.

Distance Per Stroke (DPS)

Measured in: meters (m)
Average distance traveled with each stroke cycle. Higher DPS indicates better efficiency and propulsion.
Freestyle: 1.8-2.5m | Backstroke: 1.6-2.3m | Breaststroke: 1.3-2.0m | Butterfly: 1.5-2.2m

Crossover Score

Measures hand crossover past the body midline. Lower scores indicate better technique with less energy wasted on lateral movements.
Target: Below 0.15 for good technique (all strokes)

Body Position & Alignment

Body Alignment

Measured in: degrees
Deviation from horizontal body position. A more horizontal body creates less drag and increases efficiency.
Freestyle/Backstroke: Below 10° | Breaststroke/Butterfly: 10-25° (undulation is normal)

Shoulder Roll / Body Rotation

Measured in: degrees
The amplitude of your body rotation. Essential for freestyle/backstroke, minimal for breaststroke/butterfly.
Freestyle/Backstroke: 45-60° | Breaststroke/Butterfly: 0-15° (symmetric strokes)

Efficiency & Propulsion

Average Speed

Measured in: meters per minute (m/min)
Estimated swimming speed based on body movement analysis and torso scale calculations.

SWOLF Score

Swimming efficiency metric combining time and stroke count (Time + Strokes = SWOLF). Lower is better.
Good recreational: 40-50 | Elite: Below 35

Kick Rhythm

Measured in: kicks per minute (kpm)
Frequency of leg kicks. Should be coordinated with arm strokes (typically 2-beat or 6-beat kick for freestyle).

Enhanced Biomechanical Metrics

Head Position Stability

Score: 0-100 (higher = more stable)
Tracks nose position (landmark #0) over time to measure head steadiness. Stable head position reduces drag and improves streamlining.

Hip Depth

Measures average hip position and variation using hip landmarks (#23, #24). Consistent hip depth near the surface minimizes drag.

Pull Path Efficiency

Score: 0-100 (higher = more efficient)
Tracks wrist movement through the water to measure how efficiently your hand generates propulsion. Includes catch angle measurement.

Breathing Metrics

Measures breathing angle, frequency (breaths/min), and timing consistency (0-100). Consistent breathing maintains rhythm and reduces drag disruption.

Kick Symmetry

Score: 0-100 (higher = more symmetric)
Compares left/right ankle movement patterns to detect imbalances. Asymmetric kicks cause lateral drift and waste energy.

Knee Flexion

Measured in: degrees
Three-point angle calculation (hip-knee-ankle) measuring knee bend during kicking. Proper flexion optimizes propulsion without creating excess drag.
Target: 120-160° (varies by stroke)

Propulsive Phase Duration

Measured in: percentage of stroke cycle
Analyzes wrist acceleration to identify periods of active propulsion. Higher propulsive percentage indicates more efficient swimming.
Target: 40-60% of stroke cycle

Elbow Angle During Catch

Measured in: degrees
Measures elbow flexion at the catch point for both arms. Proper early vertical forearm (EVF) maximizes catch surface area.
Target: 90-120° for high elbow catch

Underwater Speed After Start

Measures dolphin kick count, breakout timing (seconds), underwater speed (m/s), and breakout distance. Efficient underwater phases can significantly improve overall times.

PDF Export

Analysis Report Export

Download a professional PDF report of your analysis results including all core and enhanced metrics, stroke detection details, and color-coded performance ratings. Access via the results page or the API endpoint /api/jobs/{job_id}/export/pdf.
Metrics are estimates based on video analysis. Actual values may vary based on video quality, camera angle, and water conditions.

🧠 ML Model & Advanced Stroke Detection

Stroke Detection System v2.2

Swim ProHub 360 uses an advanced 5-phase biomechanical pipeline to automatically detect your swimming stroke type with high accuracy. The system analyzes body movement patterns without requiring manual training or labeled data.

How It Works: The 5-Phase Pipeline

Phase 1: Feature Extraction

27+ biomechanical features extracted from MediaPipe pose landmarks including:

  • Phase offset: Timing difference between left and right arm movements
  • Arm synchronization: Correlation between left/right arm trajectories
  • Hip rotation variance: Amount of body roll (key indicator for freestyle/backstroke)
  • Kick patterns: Flutter vs synchronized kick detection
  • Glide detection: Identifies characteristic breaststroke glide phase
  • Velocity & acceleration: Arm movement dynamics
  • Entry angle: Hand entry position relative to shoulder line

Phase 2: Temporal Consistency

Cycle detection via autocorrelation finds repeating patterns in your stroke. The system validates that detected cycles are consistent and fall within realistic timing bounds (0.8-2.5 seconds per cycle). This ensures the classification is based on actual swimming patterns, not noise.

Phase 3: Hierarchical Classification

Two-stage decision tree:

  • Stage 1: Alternating (freestyle/backstroke) vs Synchronized (breaststroke/butterfly)
    Key features: arm_sync > 0.95 AND kick_sync > 0.95 → synchronized strokes
  • Stage 2: Specific stroke identification within each category
    Uses hip rotation, shoulder roll, kick amplitude, and glide patterns

Phase 4: Viterbi HMM Smoothing

Hidden Markov Model applies temporal smoothing across detected stroke cycles. This reduces false positives from single-frame anomalies and ensures the final classification represents the dominant stroke pattern throughout the video.

Phase 5: Confidence Calibration

Multi-factor confidence scoring with four tiers:

  • ● HIGH (≥90%): Accept without review
  • ● MEDIUM (≥75%): Reliable, optional review
  • ● LOW (≥60%): Requires manual verification
  • ● REJECT (<60%): Insufficient evidence

MediaPipe Pose Detection

Body Landmark System

The analysis uses Google's MediaPipe Pose to detect 33 body landmarks per frame. Only landmarks with visibility confidence > 0.3 are used for calculations.

🏊 Upper Body
  • Wrists: #15 (left), #16 (right)
  • Elbows: #13 (left), #14 (right)
  • Shoulders: #11 (left), #12 (right)
🦵 Lower Body
  • Hips: #23 (left), #24 (right)
  • Knees: #25 (left), #26 (right)
  • Ankles: #27 (left), #28 (right)

Key Classification Rules

Definitive Synchronized Detection

If arm_sync > 0.95 AND kick_sync > 0.95, the stroke is definitively breaststroke or butterfly. This rule has near 100% accuracy for synchronized strokes.

Hip Rotation Analysis

High rotation (>0.08): Indicates alternating strokes (freestyle/backstroke)
Low rotation (<0.015): Indicates synchronized strokes (breaststroke/butterfly)

Kick Synchronization

Flutter kick (<0.55): Freestyle or backstroke
Synchronized kick (>0.90): Butterfly or breaststroke

Glide Detection

Detects characteristic breaststroke glide phases (≥3 glides, each ≥400ms duration). Unique to breaststroke among the four competitive strokes.

Model Training & Validation

Rule-Based Expert System (Not Neural Network)

Unlike traditional machine learning models, our stroke detection uses a biomechanically-grounded expert system. This approach offers several advantages:

  • No training data required: Rules derived from swimming biomechanics research
  • Fully interpretable: Every classification decision can be explained
  • Generalizes perfectly: Works on any swimmer without prior examples
  • Consistent performance: No overfitting or dataset bias
  • Real-time capable: No GPU inference needed

Scientific Validation

The classification rules are calibrated from peer-reviewed biomechanics research:

  • Seifert, L. et al. (2014) - "Inter-limb coordination and energy cost in swimming"
  • Sanders, R.H. et al. (2015) - "Body roll in freestyle swimming" - Sports Biomechanics
  • Maglischo, E.W. (2003) - "Swimming Fastest" - Human Kinetics
  • SwimXYZ Dataset - Real-world swimming video benchmark

Advanced Parameters

Detection Sensitivity Settings

The system uses research-calibrated thresholds optimized for real-world videos:

Parameter Threshold Purpose
MIN_VISIBILITY 0.3 Minimum confidence for including a landmark
MIN_FRAMES 90 frames Minimum video length (3 seconds @ 30fps)
IDEAL_FRAMES 150 frames Optimal length for highest accuracy (5 seconds)
MIN_VALID_RATIO 15% Minimum frames with valid pose detection
hip_rotation_freestyle > 0.015 High body roll indicates alternating strokes
synchronized_arm_sync > 0.95 Both arms move together (butterfly/breaststroke)
kick_sync_synchronized > 0.90 Dolphin or whip kick pattern
glide_duration_min ≥ 400ms Detects breaststroke glide phase

MediaPipe Pose Landmarks

33-Point Body Tracking

Each video frame is analyzed using Google's MediaPipe Pose model, which detects 33 body landmarks with x, y, z coordinates and visibility confidence. The system uses these landmarks to track body movement:

🏊 Arms & Shoulders
Left Wrist#15
Right Wrist#16
Left Elbow#13
Right Elbow#14
Left Shoulder#11
Right Shoulder#12

Used for: Stroke rate, phase offset, arm synchronization, entry angle, crossover detection

🦵 Legs & Hips
Left Hip#23
Right Hip#24
Left Knee#25
Right Knee#26
Left Ankle#27
Right Ankle#28

Used for: Kick rhythm, hip rotation, body alignment, kick synchronization

Feature Weights & Importance

How Features Are Prioritized

Not all features contribute equally. The system uses weighted scoring based on biomechanical significance:

Arm Alternation (4.0) - Most Important
Phase Offset (3.5)
Glide Detection (2.5)
Arm Sync (2.5)
Hip Rotation, Kick, Body Position (2.0)
Velocity & Acceleration (1.5)
Entry Angle & Elbow (1.0)

Higher weights mean the feature has stronger influence on the final stroke classification. Arm alternation and phase offset are the most reliable indicators.

Handling Real-World Video Challenges

Sparse Pose Detection

Real videos often have only 17-82% pose detection rate due to splashing, lighting, or body occlusion. The system handles this by analyzing actual sparse samples without interpolation for critical features.

Temporal Smoothing

Viterbi Hidden Markov Model smooths classifications across detected stroke cycles, filtering out single-frame anomalies while preserving genuine stroke transitions.

Adaptive Thresholds

Thresholds are calibrated for real-world video conditions, balancing sensitivity (detecting subtle patterns) with robustness (avoiding false positives from noise).
For technical details on the implementation, see src/analysis/stroke_detection.py in the source code. The system is open-source and follows established biomechanics research.
Accuracy Note: While the detection system achieves 100% accuracy on test videos with good visibility, real-world accuracy depends on video quality, camera angle, lighting, and water clarity. Always verify the detected stroke type matches your actual technique.

Video Comparison

Comparing Two Videos

The comparison feature allows you to upload two videos and analyze the differences between them. This is useful for:

  • Tracking your progress over time
  • Comparing your technique to a coach or reference video
  • Analyzing the effect of technique changes
  • Before/after comparisons for training interventions

Understanding Comparison Results

  • Overall Score: A 0-100 score indicating overall technique quality
  • Improvement Score: Positive values indicate Video 2 is better than Video 1
  • Improvements: Metrics that got better in the second video
  • Regressions: Areas where the second video showed decline
For accurate comparisons, try to record both videos from the same angle and distance, with similar lighting conditions.

AI Swimming Coach

Getting Personalized Advice

After your video is analyzed, you can chat with the AI Coach to get personalized advice about your swimming technique. The AI has access to all your metrics and can provide specific recommendations.

Example Questions to Ask

  • "How can I improve my stroke efficiency?"
  • "What does my SWOLF score mean and how can I lower it?"
  • "My body alignment seems off - what drills can help?"
  • "How do I reduce my stroke rate while maintaining speed?"
  • "What's causing my crossover issue?"
  • "Can you suggest a training plan based on my metrics?"

Correcting Analysis Errors

If you notice the AI misidentified your stroke type, you can correct it by saying something like:

  • "I was actually swimming backstroke, not freestyle"
  • "This is butterfly, please recalculate"

The AI will acknowledge the correction and provide adjusted recommendations.

Troubleshooting

Upload fails or times out
  • Check that your file is under 100MB
  • Try a different browser (Chrome recommended)
  • Check your internet connection stability
  • Try compressing the video with a tool like HandBrake
Analysis shows inaccurate metrics
  • Ensure the camera captures your full body
  • Record from the side for best results
  • Improve lighting if possible
  • Avoid excessive splashing that obscures body parts
Wrong stroke type detected
Use the AI Coach chat to correct the stroke type. The system will acknowledge the correction and adjust its recommendations accordingly.
Analysis stuck at a certain percentage
  • Analysis can take 1-5 minutes depending on video length
  • If stuck for more than 10 minutes, try refreshing and uploading again
  • Shorter videos (30 seconds to 2 minutes) process faster

Frequently Asked Questions

Is my video stored on your servers?
Videos are stored temporarily for analysis and can be deleted from the History tab. We recommend deleting sensitive videos after reviewing your results.
Can I analyze underwater footage?
Yes, but accuracy may be reduced due to light refraction in water. Above-water side-view footage typically yields the best results.
What cameras work best?
Any modern smartphone camera works well. For professional results, action cameras like GoPro provide excellent quality. 720p or higher resolution is recommended.
How accurate are the metrics?
Metrics are estimates based on computer vision analysis. While they're useful for tracking relative improvements and identifying technique issues, they may not match professional timing systems exactly.
Can I use this for competitive training?
Absolutely! The tool is designed for swimmers of all levels. Competitive swimmers can use it to track progress, identify technique flaws, and get AI-powered coaching suggestions.
Is there an API for integration?
Yes! We provide a full REST API for integration with other applications. Visit the API Documentation for details and code examples.

Need More Help?

Contact Support

If you're experiencing issues not covered in this guide, please contact our support team with:

  • A description of the issue
  • The browser and device you're using
  • Screenshots if applicable
  • Your job ID (found in the URL of your results page)
↑