WGANVO: Monocular Visual Odometry based on Generative Adversarial Networks

TL;DR

WGANVO introduces a deep learning monocular visual odometry method that estimates absolute scale without prior knowledge, trained semi-supervised, and performs in real-time with promising accuracy on KITTI dataset.

Contribution

It presents WGANVO, a novel neural network approach for monocular visual odometry that recovers absolute scale without additional information.

Findings

Operates in real-time on KITTI dataset.

Achieves encouraging accuracy in pose estimation.

Does not require prior scene scale knowledge.

Abstract

In this work we present WGANVO, a Deep Learning based monocular Visual Odometry method. In particular, a neural network is trained to regress a pose estimate from an image pair. The training is performed using a semi-supervised approach. Unlike geometry based monocular methods, the proposed method can recover the absolute scale of the scene without neither prior knowledge nor extra information. The evaluation of the system is carried out on the well-known KITTI dataset where it is shown to work in real time and the accuracy obtained is encouraging to continue the development of Deep Learning based methods.