END-TO-END DEPTH from MOTION with STABILIZED MONOCULAR VIDEOS

C. Pinard; L. Chevalley; A. Manzanera; D. Filliat

doi:10.5194/isprs-annals-IV-2-W3-67-2017

END-TO-END DEPTH from MOTION with STABILIZED MONOCULAR VIDEOS

C. Pinard
, L. Chevalley
, A. Manzanera
, D. Filliat

Parrot
ENSTA ParisTech

Research output: Contribution to journal › Conference article › peer-review

Abstract

We propose a depth map inference system from monocular videos based on a novel dataset for navigation that mimics aerial footage from gimbal stabilized monocular camera in rigid scenes. Unlike most navigation datasets, the lack of rotation implies an easier structure from motion problem which can be leveraged for different kinds of tasks such as depth inference and obstacle avoidance. We also propose an architecture for end-to-end depth inference with a fully convolutional network. Results show that although tied to camera inner parameters, the problem is locally solvable and leads to good quality depth prediction.

Original language	English
Pages (from-to)	67-74
Number of pages	8
Journal	ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Volume	4
Issue number	2W3
DOIs	https://doi.org/10.5194/isprs-annals-IV-2-W3-67-2017
Publication status	Published - 18 Aug 2017
Externally published	Yes
Event	4th ISPRS International Conference on Unmanned Aerial Vehicles in Geomatics, UAV-g 2017 - Bonn, Germany Duration: 4 Sept 2017 → 7 Sept 2017

Keywords

Dataset
Deep Learning
Depth from Motion
End-to-end
Monocular
Navigation

Access to Document

10.5194/isprs-annals-IV-2-W3-67-2017

Cite this

@article{6917b7e6f3c845be80498e2f9bcb9afb,

title = "END-TO-END DEPTH from MOTION with STABILIZED MONOCULAR VIDEOS",

abstract = "We propose a depth map inference system from monocular videos based on a novel dataset for navigation that mimics aerial footage from gimbal stabilized monocular camera in rigid scenes. Unlike most navigation datasets, the lack of rotation implies an easier structure from motion problem which can be leveraged for different kinds of tasks such as depth inference and obstacle avoidance. We also propose an architecture for end-to-end depth inference with a fully convolutional network. Results show that although tied to camera inner parameters, the problem is locally solvable and leads to good quality depth prediction.",

keywords = "Dataset, Deep Learning, Depth from Motion, End-to-end, Monocular, Navigation",

author = "C. Pinard and L. Chevalley and A. Manzanera and D. Filliat",

note = "Publisher Copyright: {\textcopyright} Authors 2017.; 4th ISPRS International Conference on Unmanned Aerial Vehicles in Geomatics, UAV-g 2017 ; Conference date: 04-09-2017 Through 07-09-2017",

year = "2017",

month = aug,

day = "18",

doi = "10.5194/isprs-annals-IV-2-W3-67-2017",

language = "English",

volume = "4",

pages = "67--74",

journal = "ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences",

issn = "2194-9042",

number = "2W3",

}

TY - JOUR

T1 - END-TO-END DEPTH from MOTION with STABILIZED MONOCULAR VIDEOS

AU - Pinard, C.

AU - Chevalley, L.

AU - Manzanera, A.

AU - Filliat, D.

N1 - Publisher Copyright: © Authors 2017.

PY - 2017/8/18

Y1 - 2017/8/18

N2 - We propose a depth map inference system from monocular videos based on a novel dataset for navigation that mimics aerial footage from gimbal stabilized monocular camera in rigid scenes. Unlike most navigation datasets, the lack of rotation implies an easier structure from motion problem which can be leveraged for different kinds of tasks such as depth inference and obstacle avoidance. We also propose an architecture for end-to-end depth inference with a fully convolutional network. Results show that although tied to camera inner parameters, the problem is locally solvable and leads to good quality depth prediction.

AB - We propose a depth map inference system from monocular videos based on a novel dataset for navigation that mimics aerial footage from gimbal stabilized monocular camera in rigid scenes. Unlike most navigation datasets, the lack of rotation implies an easier structure from motion problem which can be leveraged for different kinds of tasks such as depth inference and obstacle avoidance. We also propose an architecture for end-to-end depth inference with a fully convolutional network. Results show that although tied to camera inner parameters, the problem is locally solvable and leads to good quality depth prediction.

KW - Dataset

KW - Deep Learning

KW - Depth from Motion

KW - End-to-end

KW - Monocular

KW - Navigation

U2 - 10.5194/isprs-annals-IV-2-W3-67-2017

DO - 10.5194/isprs-annals-IV-2-W3-67-2017

M3 - Conference article

AN - SCOPUS:85030253114

SN - 2194-9042

VL - 4

SP - 67

EP - 74

JO - ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences

JF - ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences

IS - 2W3

T2 - 4th ISPRS International Conference on Unmanned Aerial Vehicles in Geomatics, UAV-g 2017

Y2 - 4 September 2017 through 7 September 2017

ER -

END-TO-END DEPTH from MOTION with STABILIZED MONOCULAR VIDEOS

Abstract

Keywords

Access to Document

Fingerprint

Cite this