Related Experiment Video
Updated: Nov 20, 2025
06:54
Photorealistic Learned Landscapes for Augmented Reality
Published on: June 27, 2025
467
A Hybrid Approach to Industrial Augmented Reality Using Deep Learning-Based Facility Segmentation and Depth
Minseok Kim1, Sung Ho Choi2, Kyeong-Beom Park2
1Korea Institute of Science and Technology Information (KISTI), Daejeon 34141, Korea.
Sensors (Basel, Switzerland)
|January 20, 2021
Summary
This study introduces a novel hybrid approach for industrial augmented reality (AR) that overcomes visual mismatches and occlusions. By using deep learning for facility segmentation and depth prediction, it enhances manufacturing tasks without markers or depth cameras.
Area of Science:
- Computer Vision
- Augmented Reality
- Industrial Automation
Background:
- Traditional augmented reality (AR) methods face challenges in industrial settings, including visual mismatches, incorrect occlusions, and limited augmentation capabilities.
- These limitations stem from the inability to estimate depth from AR images and the reliance on physical AR markers, hindering effective manufacturing task execution.
Purpose of the Study:
- To propose a hybrid augmented reality (AR) approach that complements existing methods for industrial applications.
- To address limitations of conventional AR, such as visual mismatching and occlusions, by enabling depth estimation and object relationship calculation without AR markers or depth cameras.
Main Methods:
- Employs deep learning-based instance segmentation on RGB images to extract outlines of physical objects.
- Utilizes a deep learning-based depth prediction method to estimate a depth map and 3D point cloud for detected objects.
- Calculates 3D spatial relationships among physical objects using segmented point cloud data to resolve visual and occlusion issues.
Main Results:
- The proposed hybrid AR approach effectively solves visual mismatch and occlusion problems in industrial environments.
- Demonstrates the capability to handle dynamically operating or moving facilities, such as robots, which is a limitation of conventional AR.
- Quantitative and qualitative analyses confirm the superiority of the proposed approach over existing AR methods.
Conclusions:
- The developed hybrid AR method offers a scalable and effective solution for industrial augmentation, enhancing worker performance.
- The approach's ability to handle dynamic environments and complex visual issues validates its originality and practical applicability.
- Case studies indicate broad potential for application beyond manufacturing, confirming the method's versatility.
