Accession Number:

AD1166906

Title:

Deep Learning Techniques to Estimate 3D Position in Stereoscopic Imagery

Descriptive Note:

[Technical Report, Master's Thesis]

Corporate Author:

AIR FORCE INSTITUTE OF TECHNOLOGY WRIGHT-PATTERSON AFB OH

Personal Author(s):

• Nicholson, Jonathan I

Report Date:

2022-03-24

Pagination or Media Count:

94

Abstract:

Current AAR efforts utilize machine vision algorithms to estimate the pose of a receiver aircraft. However, these algorithms are dependent on several conditions such as the availability of precise 3D aircraft models the accuracy of the pipeline significantly degrades in the absence of high-quality information given beforehand. We propose a deep learning architecture that estimates the 3D position of an object based on stereoscopic imagery. We investigate the use of both machine learning techniques and neural networks to directly regress the 3D position of the receiver aircraft. We present a new position estimation framework that is based on the differences between two stereoscopic images without relying on the stereo block matching algorithm. We analyze the speed and accuracy of its predictions and demonstrate the effectiveness of the architecture in mitigating various visual occlusions.

Descriptors:

• supervised machine learning
• computational science
• computer vision
• artificial intelligence software
• information systems
• machine learning
• neural networks
• air force
• artificial intelligence
• aircrafts
• recurrent neural networks
• tanker aircraft
• automata theory
• computer languages
• convolutional neural networks
• information science
• stereo cameras

Subject Categories:

• Cybernetics
• Photography
• Military Aircraft Operations

Distribution Statement:

[A, Approved For Public Release]