Line-Based 6-DoF Object Pose Estimation and Tracking With an Event Camera.

Area of Science:

• Computer Vision
• Robotics
• Sensor Fusion

Background:

• Object pose estimation is crucial for 3D vision applications.
• Event cameras offer advantages like high dynamic range and low latency, ideal for challenging environments.
• Standard cameras struggle with high dynamic range and fast motion, necessitating complementary sensors.

Purpose of the Study:

• To propose a robust line-based pose estimation and tracking method for planar and non-planar objects using event cameras.
• To leverage the unique properties of event cameras for improved object tracking accuracy.
• To introduce a new dataset for evaluating event-based moving object tracking.

Main Methods:

• Extracting object lines directly from event data.
• Employing a globally-optimal Branch-and-Bound approach for initial pose estimation without prior correspondences.
• Utilizing event-line matching for 2D-3D model correspondences and refining poses by minimizing event-line distances.
• Implementing robust estimation with weighted event assignments based on distances.

Main Results:

• The proposed method demonstrates robust and accurate object pose estimation and tracking.
• Validated performance against state-of-the-art methods on synthetic data and a newly created event-based dataset.
• Successful extraction of object lines and establishment of 2D-3D correspondences from raw event data.

Conclusions:

• The developed line-based method effectively utilizes event camera data for precise object pose estimation and tracking.
• The approach shows significant improvements in robustness and accuracy, particularly in challenging scenarios.
• The novel event-based dataset facilitates further research and benchmarking in this domain.