Verifying Athlete Performance Data?
Verifying Athlete Jump Heights with 90% Accuracy Using Phone Video
.avif)
Case Study Summary
Executive summary
A football sponsorship and talent-discovery platform lets fans and brands back young grassroots players early in their careers through auditable micro-sponsorship. Players record skill challenges on their phones, starting with a vertical jump, and viewers, scouts, and sponsors use those videos to decide who to support. The catch was that jump heights were self-reported, with no way to check them against the footage. Since sponsorship money followed those numbers, an unverified claim was a risk to the platform's core business.
WhizzBridge built the verification engine inside the platform. The engine measures jump height from the submitted video itself, using a reference beverage can in the frame as a ruler and a second, physics-based check on the athlete's own motion. Jump heights now come from the video instead of from the player. Across recorded jumps, the engine achieved approximately 90% accuracy, with best-performing cases within 1 cm of the actual jump height and average processing time of about 35 seconds per video on a T4 GPU.
~90%
Measurement accuracy
Across recorded jumps
1 cm
Best-case precision
Of actual jump height
~35 sec
Processing per video
On an NVIDIA T4 GPU
Challenges
- Establishing Accurate Measurement Scale: Ordinary videos have no inherent scale, making it difficult to translate pixels into actual jump height. The system needed to reliably detect a branded beverage can of known dimensions to establish a real-world measurement reference.
- Handling Unreliable Video Signals: Motion blur made foot positions unreliable for detecting takeoff and landing, while pose-tracking jitter and camera perspective distortion affected measurement accuracy.
- Controlling Processing Time and Costs: The verification engine needed to process videos efficiently while keeping third-party OCR API calls and associated costs under control as the platform scaled.
The platform needed a reliable way to verify athlete jump heights from ordinary phone videos.
.avif)
WhizzBridge’s Solution
WhizzBridge developed a video-based verification engine that combines object-based calibration with physics-based measurement to deliver more reliable jump-height estimates.
- Reference-Based Measurement: Used a branded beverage can of known height as a real-world reference to calculate athlete jump measurements from video footage.
- Physics-Based Verification: Implemented an independent measurement method based on the athlete's airborne motion, cross-checking both measurements and using physics as a fallback when the can couldn't be measured reliably.
- Improved Detection and Calibration: Redesigned detection logic, applied pose smoothing, and improved takeoff and landing detection to address motion blur, tracking jitter, and perspective distortion.
- Processing Optimization: Cached OCR results and introduced a hard limit on OCR calls to control third-party costs and processing time.
- Measurement Confidence Scoring: Added pose-confidence scores alongside jump-height measurements, allowing scouts and sponsors to assess the reliability of individual readings.
Results We Achieved
- ~90% Measurement Accuracy: Achieved approximately 90% accuracy across recorded jumps, providing video-based measurements instead of relying on athlete-reported figures.
- Within 1 cm of Actual Jump Height: Best-performing measurement cases came within 1 cm of the actual jump height.
- ~35 Seconds Processing Time: Processed each submitted video in an average of approximately 35 seconds using an NVIDIA T4 GPU.
The verification engine replaced self-reported jump heights with measurements derived directly from submitted videos, giving sponsors and scouts more reliable performance data.
Seeking a Partner? Our expert team is here for you
To get started, we would like to gather more information about your needs. We will evaluate your application and set up a free estimation call
