How AI Analyzes Your Boxing
Modern AI boxing analysis relies on a combination of computer vision and machine learning. At the core of CombatSense AI is MediaPipe Pose, a real-time pose estimation framework developed by Google.
What is Pose Estimation?
Pose estimation is the process of detecting human body keypoints — joints like wrists, elbows, and shoulders — from a video frame. MediaPipe identifies 33 landmarks on the human body, each with an (x, y, z) coordinate and a confidence score.
For boxing analysis, the most critical landmarks are:
- Wrists (left and right) — primary punch detection points
- Elbows — used to calculate arm extension
- Shoulders — reference points for movement direction
Detecting a Punch
A punch is detected when three conditions are met simultaneously:
- Speed threshold — The wrist moves faster than a minimum velocity (typically 0.08 m/s normalized)
- Arm extension — The ratio of wrist-to-shoulder distance vs. total arm length exceeds 0.88
- Visibility confidence — The landmark confidence score is above 0.7
When all three conditions are true, the system classifies the punch type based on movement direction:
- Strong upward movement → Uppercut
- Strong horizontal movement → Hook
- Forward extension with high arm extension → Cross
- Fast movement, other direction → Jab
Why This Matters for Training
Traditional boxing coaching requires a trainer present at every session. AI analysis democratizes access to performance feedback — you can train alone and still receive objective data about your technique.
The key metrics extracted from each session include total strike count, strikes per minute, average speed, and technique consistency.
Limitations and Accuracy
No AI system is perfect. Current limitations include accuracy drops in poor lighting, overlapping limbs confusing the pose estimator, and very fast combinations being partially missed.
Despite these limitations, the system achieves reliable detection for standard training sessions filmed with a modern smartphone.
The Future of AI Boxing Analysis
As pose estimation models improve and mobile hardware becomes more powerful, real-time analysis during training will become standard. The goal is not to replace coaches, but to give every boxer access to objective performance data.