Approach to Enhancing Panoramic Segmentation in Indoor Construction Sites Based on a Perspective Image Segmentation Foundation Model DOI Creative Commons
Juho Han,

Sebeen Yoon,

Mingyun Kang

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(9), P. 4875 - 4875

Published: April 27, 2025

Panoramic images in indoor construction sites are gaining attention as valuable tools for process monitoring and quality assessment. However, despite the environmental complexity demand high segmentation performance environments, scarcity of specialized models datasets has created a gap between technological advancements practical application, thus hindering effective utilization panoramic images. To address these challenges, this study proposes novel approach leveraging Segment Anything Model (SAM), perspective image foundation model, to enhance existing models. The proposed method iteratively executes SAM with adjusted input parameters extract objects varying sizes subsequently applies filtering algorithms retain valid objects. Then, label assignment merging processes performed based on predictions from target model improve accuracy. experimental was conducted using Panoplane360, specifically designed plane segmentation, model. A quantitative evaluation measure exactness assignment, two qualitative evaluations were assess whether assigned labels accurately represent actual planar information. results confirmed that significantly improves compared conventional approaches. findings highlight potential SAM-based methods accuracy dynamic environments. Furthermore, provides advantages, it without requiring additional datasets. Future research will focus resolving computational efficiency issues resulting iterative execution extend applicability diverse tasks

Language: Английский

Approach to Enhancing Panoramic Segmentation in Indoor Construction Sites Based on a Perspective Image Segmentation Foundation Model DOI Creative Commons
Juho Han,

Sebeen Yoon,

Mingyun Kang

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(9), P. 4875 - 4875

Published: April 27, 2025

Panoramic images in indoor construction sites are gaining attention as valuable tools for process monitoring and quality assessment. However, despite the environmental complexity demand high segmentation performance environments, scarcity of specialized models datasets has created a gap between technological advancements practical application, thus hindering effective utilization panoramic images. To address these challenges, this study proposes novel approach leveraging Segment Anything Model (SAM), perspective image foundation model, to enhance existing models. The proposed method iteratively executes SAM with adjusted input parameters extract objects varying sizes subsequently applies filtering algorithms retain valid objects. Then, label assignment merging processes performed based on predictions from target model improve accuracy. experimental was conducted using Panoplane360, specifically designed plane segmentation, model. A quantitative evaluation measure exactness assignment, two qualitative evaluations were assess whether assigned labels accurately represent actual planar information. results confirmed that significantly improves compared conventional approaches. findings highlight potential SAM-based methods accuracy dynamic environments. Furthermore, provides advantages, it without requiring additional datasets. Future research will focus resolving computational efficiency issues resulting iterative execution extend applicability diverse tasks

Language: Английский

Citations

0