Core Concepts
Comprehensive perception of human beings is crucial for safe human-robot interaction, achieved through a hierarchically connected tree structure integrating multi-source visual information.
Abstract
The content discusses the development of an active vision system using multiple cameras to capture RGB-D data dynamically for human-robot interaction. A hierarchically connected tree model is proposed to fuse localized visual information, enhancing keypart recognition and obstacle avoidance capabilities of robotic arms.
1. Introduction:
- Importance of comprehensive human sensing in HRI.
2. Methodology:
- Development of an active vision system with multiple cameras.
3. Hierarchical Tree Model:
- Structuring the human body as a directed tree for integration.
4. Key Part Recognition:
- Enhancing keypart recognition accuracy through multi-camera active vision.
5. Human Pose Estimation:
- Precise estimation of human body poses using the proposed method.
6. Obstacle Avoidance:
- Successful obstacle avoidance demonstrated with MCAV implementation.
Stats
RGB-Dカメラは、キーパートの認識精度を79.82%から94.73%に向上させました。
Quotes
"Utilizing RGB-D data and HRNet, the 3D positions of keypoints are analytically estimated."
"Our method enhances keypart recognition recall from 69.20% to 90.10%."