UnDeepVO: Monocular Visual Odometry through Unsupervised Deep Learning

TL;DR

UnDeepVO is an unsupervised deep learning-based monocular visual odometry system that estimates 6-DoF pose and depth, recovering absolute scale using stereo training but operating with monocular input.

Contribution

It introduces an unsupervised learning scheme for monocular VO that also recovers absolute scale, a challenge in monocular systems.

Findings

Achieves good pose accuracy on KITTI dataset

Uses stereo pairs for training but monocular images for testing

Employs a loss function based on dense spatial and temporal information

Abstract

We propose a novel monocular visual odometry (VO) system called UnDeepVO in this paper. UnDeepVO is able to estimate the 6-DoF pose of a monocular camera and the depth of its view by using deep neural networks. There are two salient features of the proposed UnDeepVO: one is the unsupervised deep learning scheme, and the other is the absolute scale recovery. Specifically, we train UnDeepVO by using stereo image pairs to recover the scale but test it by using consecutive monocular images. Thus, UnDeepVO is a monocular system. The loss function defined for training the networks is based on spatial and temporal dense information. A system overview is shown in Fig. 1. The experiments on KITTI dataset show our UnDeepVO achieves good performance in terms of pose accuracy.