New framework enables calibration-free 3D multi-camera people tracking

Researchers have developed a novel calibration-free framework for 3D multi-camera people tracking in indoor environments. This system integrates multiple deep learning models, including YOLOX for detection, BoT-SORT for tracking, OsNet for appearance embedding, and a Visual Geometry Grounded Transformer for geometric reconstruction. By inferring the 3D structure directly from visual data and using a pose-guided 3D lifting strategy, the framework eliminates the need for precise camera calibration, a significant bottleneck in previous methods. Evaluations on the AI City Challenge 2024 demonstrated a competitive HOTA score without ground-truth calibration data. AI

IMPACT This research could streamline the setup and deployment of multi-camera tracking systems by removing the need for manual calibration.

RANK_REASON This is a research paper detailing a new technical framework for computer vision. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• AI City Challenge 2024
• BoT-SORT
• Dominique Vaufreydaz
• Mmpose
• OsNet
• Visual Geometry Grounded Transformer
• Yolox Object Detection

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →