[專案介紹] - 運用空間模型及深度學習分析工程進度

To date, many construction sites rely on conventional daily construction reports for progress monitoring. Although these reports contain several necessary information for a construction site's day-to-day operations, it fails to provide a visual verification for progress status timely. As these reports are made manually, it is often labor-intensive and prone to human errors. The exponential growth of on-site visual data and the advent of computer vision techniques have created a unique opportunity to improve the vision-based automated construction progress monitoring methods. Vision-based progress monitoring methods are broadly classified into occupancy-based and appearance-based methods. To date, these methods are capable of reporting the progress of a building element in terms of binary function leveraging 4D BIM models with LOD 300-350. However, for better schedule control and micro-level monitoring, it is necessary to report the partial completion of an element and tasks associated with elements that may not be modeled in BIM. This research proposes a novel framework for computing and reporting the partial progress in terms of completion percentage using the on-site visual data, 4D BIM, and deep learning-based computer vision algorithms. The framework can automatically calculate the percentage completion for each element and task with associated elements that are not modeled based on their respective unit of measurement leveraging geometry modeling and appearance detection. The proposed framework is applied to a building construction project, and the preliminary results demonstrate its applicability to generate completion percentage per task in the lookahead schedule for accurate daily progress report generation.