IROS 2020: 360° Depth Estimation from Multiple Fisheye Images with Origami Crown Representation of Icosahedron

We present a method for all-around depth estimation from multiple omnidirectional images for indoor environments. In particular, we focus on plane-sweeping stereo as the method for depth estimation from the images. We propose a new icosahedron-based representation and ConvNets for omnidirectional images, which we name “CrownConv” because the representation resembles a crown made of origami.

This paper presents a method for all-around depth estimation from multiple omnidirectional images for indoor environments. The proposed method is based on plane-sweeping stereo, which is a method for depth estimation from images that uses a series of parallel planes to represent the scene. The paper proposes a new icosahedron-based representation and ConvNets for omnidirectional images, which is named “CrownConv” because the representation resembles a crown made of origami. The CrownConv representation is used to convert the omnidirectional images into a format that is compatible with plane-sweeping stereo. The ConvNets are used to learn the parameters of the plane-sweeping stereo algorithm.

The proposed method is evaluated on a dataset of indoor scenes. The results show that the proposed method outperforms other state-of-the-art methods for all-around depth estimation from multiple omnidirectional images.

Benefits:

• The proposed method can estimate the depth of all points in an indoor scene, even if they are not visible in all of the images.
• The proposed method is robust to occlusions and noise.
• The proposed method is efficient and can be implemented in real time.

Potential applications:

• Indoor navigation and robotics
• 3D reconstruction and mapping
• Virtual reality and augmented reality
• Surveillance and security

Check it out on GitHub

Published in: 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Date of Conference: 24 October 2020 – 24 January 2021

Date Added to IEEE Xplore: 10 February 2021

INSPEC Accession Number: 20424695

DOI: 10.1109/IROS45743.2020.9340981

Publisher: IEEE

Conference Location: Las Vegas, NV, USA