REGROUP: A Robot-Centric Group Detection and Tracking System
summary
REGROUP is an ego-centric group detection and tracking system that runs real-time on a single RGB sensor mounted on a mobile robot (ZED stereo on a Double Telepresence robot). It uses YOLO for pedestrian detection, person re-identification CNNs for appearance-based data association, and graph-based adjacency plus DFS to form groups. Benchmarks show up to 40% higher precision and 18% higher recall for detection and up to 66% higher tracking accuracy versus prior methods.
AI-written summary · about
chapters (10)
- 0:00Introduction and motivation
- 1:10Challenges in group perception
- 2:14REGROUP system overview
- 3:00Pedestrian detection and tracking
- 3:42Ego-centric group detection
- 5:07Group re-identification and tracking
- 6:09Dataset and experimental setup
- 6:50Comparison with state-of-the-art
- 8:20Tracking accuracy and precision
- 9:01Future applications and contributions
related talks
06mMultimodal Joint Head Orientation Estimation in Interacting Groups via Proxemics and Interaction ...
26mFrom Panoramic Vision to Lightweight 3D Worlds for Semantic Visual Tracking
10mGO-Finder: A Registration-Free Wearable System for Assisting Users in Finding Lost Objects via ...
21mVisualizing robot minds, and other fun stuff
13mEnhancing Object Detection & Avoidance Systems for Robots in Challenging Environments Using AI
19mEveryday Eye Contact Detection Using Unsupervised Gaze Target Discovery
27mAnalysis of sports scenes using computer vision models
15mRobot Hand Driven by Computer Vision and Reinforcement Learning