Mesh-Aware Epipolar Matching for Multi-View Multi-Person 3D Pose Estimation in Basketball

2026-05-28 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

Mesh-Aware Epipolar Matching (MAEM) is a novel, training-free framework designed for multi-view multi-person 3D pose estimation in challenging team sports like basketball. Addressing issues such as player occlusions, appearance similarities from uniforms, and limited annotated multi-view data, MAEM leverages a monocular 3D human mesh recovery model as its initial stage. It then implements a two-stage epipolar matching strategy, utilizing the recovered mesh outputs. This framework integrates disjoint-set-union-based clustering with per-joint triangulation to ensure robust cross-view association and precise 3D pose reconstruction. Experiments on public basketball datasets, SportCenter EPFL and Human-M3 Basketball, show MAEM consistently outperforms existing training-free association baselines. It achieves MPJPE/PA-MPJPE scores of 59.8/40.7 mm and 74.0/51.8 mm respectively, demonstrating the effectiveness of dense mesh geometry for association without requiring target-domain training or fine-tuning.

Key takeaway

For Computer Vision Engineers developing multi-person 3D pose estimation systems in challenging environments like sports, consider integrating training-free approaches. Your team can achieve robust cross-view association and accurate 3D pose reconstruction by leveraging dense mesh geometry from monocular recovery models. This method, exemplified by MAEM's 59.8/40.7 mm MPJPE scores, avoids extensive data annotation and domain-specific training, streamlining development and deployment.

Key insights

Dense mesh geometry significantly enhances training-free multi-view 3D pose estimation in complex sports environments.

Principles

• Training-free methods can surpass learning-based limits.
• Mesh recovery improves cross-view association.
• Epipolar geometry is robust for 3D reconstruction.

Method

MAEM uses a monocular 3D human mesh recovery frontend, followed by a two-stage epipolar matching strategy combining disjoint-set-union clustering and per-joint triangulation for 3D pose.

In practice

• Apply monocular mesh recovery for initial 3D estimates.
• Use epipolar matching for robust multi-view association.
• Evaluate training-free methods for specific domains.

Topics

• Multi-View 3D Pose Estimation
• Human Mesh Recovery
• Epipolar Geometry
• Computer Vision
• Sports Analytics
• Training-Free Methods

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• E-3DPSM: A State Machine for Event-Based Egocentric 3D Human Pose Estimation — E-3DPSM improves egocentric 3D human pose estimation by aligning continuous motion with event camera dynamics.
• Multi-scale interaction network for stereo image super-resolution — A multi-scale interaction network enhances stereo image super-resolution by optimizing intra-view and cross-view information exploitation.
• Articulation in Prime: Primitive-Based Articulated Object Understanding from a Single Casual Video — A new framework uses geometric primitive fitting to robustly recover 3D kinematics of articulated objects from single videos.
• CHOIR: Contact-aware 4D Hand-Object Interaction Reconstruction — CHOIR reconstructs 4D hand-object interactions from monocular video, explicitly using contact as a coupling signal for improved consistency.
• Stream3D: Sequential Multi-View 3D Generation via Evidential Memory — Stream3D uses an evidential memory and evidence score mechanism to enable consistent 3D generation from sequential multi-view inputs without retraining existing models.
• PanoPlane: Plane-Aware Panoramic Completion for Sparse-View Indoor 3D Gaussian Splatting — Panoramic scene completion with layout-anchored attention steering significantly improves sparse-view 3D reconstruction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.