Self-Supervised Scale Recovery for Monocular Depth and Egomotion
Estimation
B. Wagstaff, and J. Kelly. (2020)Manuscript submitted to the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2021).
Abstract
The self-supervised loss formulation for jointly training depth and egomotion
neural networks with monocular images is well studied and has demonstrated
state-of-the-art accuracy. One of the main limitations of this approach,
however, is that the depth and egomotion estimates are only determined up to an
unknown scale. In this paper, we present a novel scale recovery loss that
enforces consistency between a known camera height and the estimated camera
height, generating metric (scaled) depth and egomotion predictions. We show
that our proposed method is competitive with other scale recovery techniques
that have more information available. Further, we demonstrate how our method
facilitates network retraining within new environments, whereas other
scale-resolving approaches are incapable of doing so. Notably, our egomotion
network is able to produce more accurate estimates than a similar method that
only recovers scale at test time.
%0 Generic
%1 2020-wagstaff
%A Wagstaff, Brandon
%A Kelly, Jonathan
%D 2020
%K ambiguity camera estimation height median monocular recovery scale
%T Self-Supervised Scale Recovery for Monocular Depth and Egomotion
Estimation
%X The self-supervised loss formulation for jointly training depth and egomotion
neural networks with monocular images is well studied and has demonstrated
state-of-the-art accuracy. One of the main limitations of this approach,
however, is that the depth and egomotion estimates are only determined up to an
unknown scale. In this paper, we present a novel scale recovery loss that
enforces consistency between a known camera height and the estimated camera
height, generating metric (scaled) depth and egomotion predictions. We show
that our proposed method is competitive with other scale recovery techniques
that have more information available. Further, we demonstrate how our method
facilitates network retraining within new environments, whereas other
scale-resolving approaches are incapable of doing so. Notably, our egomotion
network is able to produce more accurate estimates than a similar method that
only recovers scale at test time.
@misc{2020-wagstaff,
abstract = {The self-supervised loss formulation for jointly training depth and egomotion
neural networks with monocular images is well studied and has demonstrated
state-of-the-art accuracy. One of the main limitations of this approach,
however, is that the depth and egomotion estimates are only determined up to an
unknown scale. In this paper, we present a novel scale recovery loss that
enforces consistency between a known camera height and the estimated camera
height, generating metric (scaled) depth and egomotion predictions. We show
that our proposed method is competitive with other scale recovery techniques
that have more information available. Further, we demonstrate how our method
facilitates network retraining within new environments, whereas other
scale-resolving approaches are incapable of doing so. Notably, our egomotion
network is able to produce more accurate estimates than a similar method that
only recovers scale at test time.},
added-at = {2021-07-06T17:14:11.000+0200},
author = {Wagstaff, Brandon and Kelly, Jonathan},
biburl = {https://www.bibsonomy.org/bibtex/246db4837e07345ab4e09b3323b5deea2/pkoch},
interhash = {0ae7d97bd3cf0887a18945bc38a3c87e},
intrahash = {46db4837e07345ab4e09b3323b5deea2},
keywords = {ambiguity camera estimation height median monocular recovery scale},
note = {Manuscript submitted to the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2021)},
timestamp = {2021-07-15T14:00:30.000+0200},
title = {Self-Supervised Scale Recovery for Monocular Depth and Egomotion
Estimation},
year = 2020
}