AI-Assisted Reuse-Oriented Building Inventory for Sustainable Construction Using Panoramic Indoor Imagery

Sammanfattning

This project investigated how artificial intelligence-based object detection could support reuse-oriented building inventory processes using panoramic and 360-degree indoor imagery. The project focused on designing and evaluating a prototype system capable of detecting reusable furnishing elements and generating structured inventory-related outputs from indoor image data. The prototype was implemented using the You Only Look Once version 8 object detection framework together with Python-based machine learning tools. The work included dataset preparation, manual annotation, transfer learning, model fine-tuning, and experimental evaluation. Initial experiments were performed using a publicly available indoor dataset, followed by fine-tuning using a custom panoramic and 360-degree indoor dataset collected with an Insta360 ONE X2 camera. The experimental results showed that the trained model was capable of detecting and classifying multiple furnishing objects within panoramic indoor environments despite image distortion and complex viewing perspectives. The best-performing fine-tuning stage achieved a precision of 64.0%, a recall of 73.9%, and an F1-score of 68.6% on the custom testing dataset. To further assess model robustness, five-fold cross-validation was performed on the custom dataset. The cross-validation experiments achieved an average precision of 68.3%, recall of 88.5%, and F1-score of 76.9%, indicating relatively stable performance across different training and validation splits. The results demonstrated that object detection methods could support automated reuse-oriented inventory workflows by generating structured detections from indoor image data. Although the developed prototype did not represent a fully automated industrial system, the project showed that computer vision methods have practical potential for improving reuse-oriented inventory processes within sustainable construction and renovation environments.