Frame-prior-constrained extrinsic refinement for stereo event camera calibration in a coaxial event-frame system

Area of Science:

• Computer Vision
• Robotics
• Sensor Fusion

Background:

• Accurate stereo extrinsic calibration is crucial for metric 3D reconstruction and motion tracking with event cameras.
• Challenges include sparse event data, feature localization errors, and baseline/pose bias, leading to amplified triangulation errors.

Purpose of the Study:

• To propose a frame-prior-constrained extrinsic refinement method for coaxial event-frame stereo vision systems.
• To improve the metric accuracy and geometric consistency of event-stereo 3D tracking.

Main Methods:

• Utilized a beam-splitter optical configuration for coaxial event and frame cameras.
• Introduced calibrated frame-stereo extrinsics as geometric priors for event-stereo extrinsic refinement.
• Optimized event-stereo rotation and translation using local 6-DoF perturbations on the Lie group manifold.

Main Results:

• Refined an initial event-stereo baseline of 312.54 mm to 316.32 mm, closely matching the physical baseline of 316.5 mm.
• Demonstrated improved 3D tracking accuracy in subsequent measurement experiments.
• Validated enhanced metric accuracy and geometric consistency of event-stereo 3D tracking.

Conclusions:

• The proposed frame-prior-constrained refinement effectively addresses challenges in event-stereo calibration.
• The method significantly enhances the reliability and precision of 3D reconstruction and motion tracking using event-stereo systems.