DeepAVO: Efficient Pose Refining with Feature Distilling for Deep Visual Odometry

TL;DR

DeepAVO introduces a novel deep learning framework with feature distilling and attention mechanisms to improve monocular visual odometry accuracy, outperforming existing methods in diverse scenarios.

Contribution

The paper proposes a four-branch CNN architecture with channel-spatial attention for enhanced motion-specific feature extraction in monocular VO.

Findings

DeepAVO outperforms state-of-the-art monocular VO methods.

Achieves competitive results with stereo VO algorithms.

Demonstrates strong generalization across outdoor and indoor datasets.

Abstract

The technology for Visual Odometry (VO) that estimates the position and orientation of the moving object through analyzing the image sequences captured by on-board cameras, has been well investigated with the rising interest in autonomous driving. This paper studies monocular VO from the perspective of Deep Learning (DL). Unlike most current learning-based methods, our approach, called DeepAVO, is established on the intuition that features contribute discriminately to different motion patterns. Specifically, we present a novel four-branch network to learn the rotation and translation by leveraging Convolutional Neural Networks (CNNs) to focus on different quadrants of optical flow input. To enhance the ability of feature selection, we further introduce an effective channel-spatial attention mechanism to force each branch to explicitly distill related information for specific Frame to Frame (F2F) motion estimation. Experiments on various datasets involving outdoor driving and indoor walking scenarios show that the proposed DeepAVO outperforms the state-of-the-art monocular methods by a large margin, demonstrating competitive performance to the stereo VO algorithm and verifying promising potential for generalization.