Augmented Reality For Indoor Navigation In A Warehouse Setting: Examining Supervised Learning-Based Pretrained CNNs for Feature Extraction in Visual Odometry
Author
Term
4. term
Publication year
2023
Submitted on
2023-05-30
Pages
53
Abstract
This report examines the potential use of Augmented Reality (AR) technol- ogy in improving warehouse manage- ment processes, specifically in tasks such as picking, receiving, and re- locating items. The report explores the technical aspects of AR and the challenges involved in developing an AR solution for warehouse naviga- tion, including the use of Simul- taneous Localization and Mapping (SLAM) algorithms for accurate posi- tional tracking. The report proposes a deep learning-based feature extrac- tor for visual odometry in large indus- trial environments, which combines the FAST key point extractor with pre-trained CNNs to generate feature descriptors and evaluates its perfor- mance against traditional feature ex- traction techniques such as ORB. The test results show that the ORB al- gorithm demonstrated superior effi- ciency and accuracy compared to the deep learning-based methods. Future work includes investigating the im- pact of extracting features from differ- ent layers within pre-trained networks and exploring the benefits of leverag- ing GPUs for computation.
This report examines the potential use of Augmented Reality (AR) technol- ogy in improving warehouse manage- ment processes, specifically in tasks such as picking, receiving, and re- locating items. The report explores the technical aspects of AR and the challenges involved in developing an AR solution for warehouse naviga- tion, including the use of Simul- taneous Localization and Mapping (SLAM) algorithms for accurate posi- tional tracking. The report proposes a deep learning-based feature extrac- tor for visual odometry in large indus- trial environments, which combines the FAST key point extractor with pre-trained CNNs to generate feature descriptors and evaluates its perfor- mance against traditional feature ex- traction techniques such as ORB. The test results show that the ORB al- gorithm demonstrated superior effi- ciency and accuracy compared to the deep learning-based methods. Future work includes investigating the im- pact of extracting features from differ- ent layers within pre-trained networks and exploring the benefits of leverag- ing GPUs for computation.
Documents