Revisiting Matching Response and Swept Feature Volumes for Wide-baseline Omnidirectional Stereo

2026-07-13

AI Summary

This work addresses the inaccuracy of confidence estimation caused by matching ambiguities in wide-baseline omnidirectional stereo vision. The authors propose an end-to-end method that requires neither additional modules nor multiple inference passes. By interpreting the matching responses from a 3D encoder-decoder as a probability distribution, the method uses its expected value as an intrinsic confidence signal. A sweep-based feature voxel resampling mechanism is introduced to jointly predict depth and surface normals, thereby enhancing geometric consistency. Integrating 3D CNN–based matching response analysis, expectation-driven confidence modeling, and 2D CNN–based meta-information prediction, the approach significantly improves both confidence estimation accuracy and normal quality while maintaining computational efficiency suitable for deployment in autonomous driving scenarios.

Abstract

In this paper, we propose a training strategy for confidence estimation in omnidirectional stereo, targeting the ambiguous matches that frequently occur in wide-baseline setups. Reinterpreting the matching responses produced by the 3D encoder decoder block, we show that their expectation values provide intrinsic confidence signals. Building on this, our method directly penalizes ambiguous responses without auxiliary heads, multi-pass inference, or additional modules, resulting in more efficient and generalized predictions. Beyond confidence, we introduce swept feature volume resampling, where response features produced by 3D CNNs are resampled using regressed positive matching indices and then processed by 2D CNNs to predict meta-information such as surface normals. This joint learning introduces auxiliary geometric regularization and improves depth coherence by leveraging additional contextual cues during response aggregation stage. Experimental results demonstrate that our approach enhances both confidence estimation and surface normal prediction while maintaining deployment practicality for autonomous mobility applications.

Problem

• wide-baseline stereo
• omnidirectional stereo
• ambiguous matches
• confidence estimation
• depth coherence

Innovation

• confidence estimation
• omnidirectional stereo
• swept feature volume
• matching response
• geometric regularization