Smart Boxing Analytics

AI-powered multi-view boxing analytics system using Computer Vision, Pose Estimation, Tracking, Punch Detection, Tactical Modeling, and Sequential Strategy Analysis.

Explore Project

About The System

boxing

The Smart Boxer Analytics System is an advanced AI-driven sports analytics platform developed for boxing performance analysis using synchronized top-view and side-view cameras.

The system performs boxer tracking, pose estimation, punch classification, target zone detection, strategy inference, probabilistic sequence modeling, and occlusion handling using multi-view fusion.

The project integrates YOLO pose estimation, segmentation, Markov models, Gaussian HMMs, temporal motion analysis, and contextual sequence prediction to generate tactical insights for coaches and athletes.

Core Features

Multi-View Tracking

Integrated top-view and side-view camera synchronization for robust boxer identification under occlusion and detection failure.

Punch Classification

Detects Left/Right Hook, Jab, Cross, and Uppercut using shoulder-elbow-wrist kinematic analysis.

Target Zone Detection

Identifies whether punches target the head or body using opponent pose keypoints and spatial mapping.

Guard Detection

Detects defensive guard posture based on wrist-to-face spatial relationships.

Strategy Modeling

Infers attacking, neutral, and defensive tactical states using motion, distance, punch tempo, and contextual behavior.

Markov Prediction Engine

Uses opponent-conditioned probabilistic sequence modeling to predict next punch and strategic transitions.

Analytics Pipeline

1. Video Acquisition

Allied Vision cameras capture synchronized top-view and side-view sparring sessions.

2. Detection & Tracking

YOLO Pose + Segmentation + ByteTrack perform real-time boxer tracking and identity consistency.

3. Pose Analysis

Extracts skeletal keypoints for punch mechanics, guard posture, and engagement analysis.

4. Punch Analytics

Punches are classified using arm trajectories, wrist velocity, and elbow angles.

5. Tactical Intelligence

Markov Models and sequence learning infer strategy transitions and opponent-conditioned actions.

6. Visualization

Real-time overlays, punch counters, tactical indicators, and analytics dashboards are generated.

Technology Stack

Python
YOLOv8
OpenCV
DeepSORT
ByteTrack
PyTorch
Pose Estimation
Markov Models
Gaussian HMM
Computer Vision
Multi-View Fusion
Sports Analytics

Research Outcomes

Robust Occlusion Handling

Improved punch recognition reliability using top-view + side-view fusion under referee and boxer occlusions.

Sequence Prediction

Opponent-conditioned Markov models achieved promising next-punch prediction performance.

Strategy Detection

Offensive, defensive, and neutral tactical states inferred from continuous movement behavior.

Coach-Centric Analytics

Converts raw training videos into actionable tactical insights for athlete performance optimization.

Project Information

Research Domain

Computer Vision, Sports Analytics, Human Pose Estimation, Tactical AI.

Institution

Indian Institute of Technology Madras (IIT Madras)

Application

Elite boxing performance analysis, training analytics, tactical coaching assistance, and combat sports AI.