Anti-reflective stereo VIO front end proposed, measured to improve trajectory accuracy

Uihyun Hwang and other authors propose a lightweight reflection-aware front end for feature-based stereo visual-inertial odometry (VIO). It uses a Kalman filter to estimate ground height and only rejects reflective virtual feature candidates whose inferred 3D positions are below the estimated ground. The method restricts the validation range with a ground-region mask, constructs frame-level ground-height measurements from stereo height uncertainty and pose uncertainty, then stabilizes the estimate with a scalar Kalman filter with NIS gating, and performs selective rejection with a safety margin set according to uncertainty. The authors report that experiments on real indoor sequences with reflective floors show that, in reflection-dominated environments, the front end overall improves trajectory accuracy while maintaining accurate ground-height estimation and being able to recover from actual ground-height changes.

AI 根据报道生成 · 6 days ago更新