Methodology
How FastBowlingLab analyses a delivery
A transparent overview of the video, pose, event-detection and reporting pipeline, and the limits of measurements made from phone footage.
Published
1. Check that the footage is usable
Before generating an analysis, the pipeline checks whether the clip plausibly contains a bowling delivery and whether a bowler can be tracked. Geometry then determines whether a particular event or measurement is observable. A camera-angle label helps choose the best available view; it does not make an off-plane measurement valid.
2. Track the athlete, not the frame
Our AI video analysis models follow the athlete through the run-up and estimate body landmarks on each usable frame. Measurements are normalized to the athlete’s body scale rather than the overall image height, reducing the effect of a bowler moving closer to the camera or occupying a different fraction of the frame.
3. Locate the events that anchor a delivery
The analysis identifies back-foot contact, front-foot contact and release before reading event-specific measurements. Release is tied to the bowling wrist relative to its own shoulder, rather than simply choosing the wrist’s highest point on screen. Foot contacts are selected as the delivery plants nearest the relevant anchors, not an arbitrary earlier run-up footfall.
4. Return evidence with the interpretation
The athlete receives artefacts that can be checked rather than a score alone:
- An annotated video with one output frame for every source frame.
- Key frames for the important delivery moments.
- Measurements and the clip used as the source for each one.
- A written analysis and four-week plan for an Analysis, or after a Pose-only Analysis is explicitly unlocked.
Research and validation sources
Model-estimated body landmarks do not by themselves establish the accuracy of a cricket measurement. Cricket-specific markerless validation and broader two-dimensional video research are therefore used to set conservative boundaries around camera perspective, calibration and joint-angle interpretation.
- Abraham, Feros & Fox (2025). Exploring the accuracy of OpenCap for three-dimensional analysis of cricket bowling. 10.1177/17479541251348081
- Shishov, Elabd, Komisar, Chong & Robinovitch (2021). Accuracy of Kinovea software in estimating body segment movements during falls captured on standard video: Effects of fall direction, camera perspective and video calibration technique. 10.1371/journal.pone.0258923
What the system does not claim
- It does not issue an official elbow-extension or bowling-legality verdict.
- It does not diagnose injury or prescribe medical treatment.
- It does not turn an obscured or off-plane joint into a confident measurement.
- It does not treat one idealized technique as correct for every athlete.