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

Accurate stereo extrinsic calibration is essential for metric 3D reconstruction and motion tracking with event cameras. However, event-stereo calibration is often affected by sparse event observations, feature-localization uncertainty, and residual baseline or pose bias, which can be amplified during triangulation. This paper proposes a frame-prior-constrained extrinsic refinement method for a coaxial event-frame stereo vision system. The system uses a beam-splitter-based optical configuration in which each event camera and frame camera share an approximately common optical path. Based on this configuration, the calibrated frame-stereo extrinsics are introduced as geometric priors to regularize the event-stereo extrinsic refinement. The event-stereo rotation and translation are optimized by applying a local 6-DoF perturbation on the Lie group manifold. The physical baseline of the experimental stereo rig is approximately 316.5 mm, whereas the initial event-stereo baseline estimated from conventional event-camera calibration is 312.54 mm. After refinement, the event-stereo baseline is corrected to 316.32 mm, and improved 3D tracking accuracy is demonstrated in subsequent measurement experiments. These results demonstrate that the proposed frame-prior-constrained refinement improves the metric accuracy and geometric consistency of event-stereo 3D tracking.