Graph Convolutional Network for 3D Object Pose Estimation in a Point Cloud.

Area of Science:

• Computer Vision
• Machine Learning
• Graph Neural Networks

Background:

• Graph Neural Networks (GNNs) learn node and edge representations.
• Graph Convolutional Networks (GCNs) use Convolutional Neural Networks (CNNs) for graph-based data.
• 3D object detection and pose estimation are crucial in computer vision.

Purpose of the Study:

• Propose a one-stage GCN approach for 3D object detection and pose estimation.
• Structure non-linearly distributed 3D points into a graph for analysis.
• Enhance bounding box estimation and object pose determination.

Main Methods:

• Utilize a one-stage GCN for processing spatially structured graph data.
• Implement a keypoint attention mechanism to aggregate relative point features.
• Employ quaternion rotation to avoid gimbal lock in 3D pose estimation.

Main Results:

• Demonstrated improved memory usage and efficiency through graph-based point feature aggregation.
• Achieved comparable performance against state-of-the-art systems in 3D object detection and pose estimation.
• Successfully estimated object category and pose with nine degrees of freedom.

Conclusions:

• The proposed GCN approach effectively handles 3D point cloud data for object detection and pose estimation.
• Graph structuring and attention mechanisms enhance feature aggregation and analysis.
• The method offers an efficient and competitive solution for 3D perception tasks.