Scholarlink

This paper presents an automated comprehensive image analysis pipeline designed for the spatially separated detection and hierarchical clustering of objects within multiple images taken for different kinds of documentation. While the original application derives from German crime scene documentation, the presented concept describes the application for various fields and use cases containing hierarchical structures of objects. In civil engineering and construction documentation, the as-built documentation, and the automated evaluation of images of construction diaries or issue documentation are useful use cases, improving the way data from multiple images in construction documentation is analyzed and interpreted. The presented approach closes the gap in existing solutions relying on additional information like BIM-models or specific photogrammetry-usages to gather (building) object relations. The parts of the pipeline are Single Image Analysis, Image Comparison, and Evaluation of Image Comparison. The Single Image Analysis includes the methods Metadata Extraction, computer vision methods such as object detection and -segmentation, a QR code reader as well as an image-rating of probability of being an overview image. The Image Comparison is executed for all image pairs and allows for the comparison of different types of information between two images and calculates the affinity-score (based on probability) for matching information for every single one of the following single comparison-methods: Metadata Comparison, Feature Matching, Object Segmentation Comparison (also combined with Feature Matching) and QR Code Comparison. The Image Comparison is done by comparing all image pairs with all available methods or more intelligently by using smarter resource-saving approaches described in the paper. The Evaluation of Image Comparison sorts the detected objects in the images hierarchically based on the previous steps. Additionally, the images get sorted room-wise into clusters. As a proof of concept, a demonstrative application with the use case of crime scene documentation is introduced, and the results are discussed.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated Hierarchical Object Clustering for Multi-Image Documentation: A Comprehensive Image Analysis Pipeline

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Michael Disser,
  • Maximilian Gehring,
  • Uwe Rüppel

摘要

This paper presents an automated comprehensive image analysis pipeline designed for the spatially separated detection and hierarchical clustering of objects within multiple images taken for different kinds of documentation. While the original application derives from German crime scene documentation, the presented concept describes the application for various fields and use cases containing hierarchical structures of objects. In civil engineering and construction documentation, the as-built documentation, and the automated evaluation of images of construction diaries or issue documentation are useful use cases, improving the way data from multiple images in construction documentation is analyzed and interpreted. The presented approach closes the gap in existing solutions relying on additional information like BIM-models or specific photogrammetry-usages to gather (building) object relations. The parts of the pipeline are Single Image Analysis, Image Comparison, and Evaluation of Image Comparison. The Single Image Analysis includes the methods Metadata Extraction, computer vision methods such as object detection and -segmentation, a QR code reader as well as an image-rating of probability of being an overview image. The Image Comparison is executed for all image pairs and allows for the comparison of different types of information between two images and calculates the affinity-score (based on probability) for matching information for every single one of the following single comparison-methods: Metadata Comparison, Feature Matching, Object Segmentation Comparison (also combined with Feature Matching) and QR Code Comparison. The Image Comparison is done by comparing all image pairs with all available methods or more intelligently by using smarter resource-saving approaches described in the paper. The Evaluation of Image Comparison sorts the detected objects in the images hierarchically based on the previous steps. Additionally, the images get sorted room-wise into clusters. As a proof of concept, a demonstrative application with the use case of crime scene documentation is introduced, and the results are discussed.