ORB-SfMLearner: ORB-Guided Self-supervised Visual Odometry with Selective Online Adaptation

TL;DR

ORB-SfMLearner introduces ORB features and selective online adaptation to improve the accuracy and generalizability of deep visual odometry, demonstrating superior performance on standard datasets.

Contribution

The paper presents a novel ORB-guided self-supervised visual odometry method with a cross-attention mechanism and online adaptation for enhanced robustness and domain generalization.

Findings

Outperforms state-of-the-art methods on KITTI and vKITTI datasets.

Uses ORB features for more robust ego-motion estimation.

Enables rapid domain adaptation through selective online tuning.

Abstract

Deep visual odometry, despite extensive research, still faces limitations in accuracy and generalizability that prevent its broader application. To address these challenges, we propose an Oriented FAST and Rotated BRIEF (ORB)-guided visual odometry with selective online adaptation named ORB-SfMLearner. We present a novel use of ORB features for learning-based ego-motion estimation, leading to more robust and accurate results. We also introduce the cross-attention mechanism to enhance the explainability of PoseNet and have revealed that driving direction of the vehicle can be explained through the attention weights. To improve generalizability, our selective online adaptation allows the network to rapidly and selectively adjust to the optimal parameters across different domains. Experimental results on KITTI and vKITTI datasets show that our method outperforms previous state-of-the-art deep visual odometry methods in terms of ego-motion accuracy and generalizability. Code is available at https://github.com/PeaceNeil/ORB-SfMLearner