Kurume Institute of Technology Simplifying and Automating Concrete Panel Inspection Using AI Haruto Esaki Project Mentor(s): Mariko Oda
In the construction industry, achieving operational efficiency through automation has become an urgent priority due to labor shortages and increasing pressure to reduce production costs. This research focuses on the development and evaluation of an automated inspection system for large-scale precast concrete panels using 2D image analysis. Traditionally, these inspections have relied on manual visual checks and specialized tools, which are labor-intensive and present challenges in maintaining digital evidence for objective quality control. To address these issues, this study evaluates an automated system that captures high-resolution images in sections and stitches them together. By applying the latest AI models, such as DepthAnythingV2 for monocular depth estimation, the system successfully identifies product boundaries even under complex factory lighting. Experimental results showed a measurement error of 35 mm. While this does not yet meet the strict field requirement of a 5 mm tolerance, the system successfully performed the automated detection and counting of embedded parts. Moving forward, we will evaluate a 3D data-driven approach using LiDAR scanners and other 3D scanning technologies to conduct a comparative performance analysis with our 2D method. By analyzing the advantages and disadvantages of each approach—including measurement accuracy, ease of data collection, and environmental constraints—this research aims to identify the most practical data format to achieve a reliable, automated inspection system for the industry. Presentation Type: Pre-Recorded Presentation (https://www.youtube.com/@cwusource5518) Keywords: Artificial Intelligence, Construction Industry, Precast Concrete Panels, Automated, 2D Image Analysis SOURCE Form ID: 9N Dialysis is an essential medical procedure for blood purification; however, it carries a severe risk of accidental needle removal, which can lead to rapid blood loss of 150 to 200 milliliters per minute. Data indicates that these accidents remain a persistent challenge, with 152 cases recorded in 2021 alone. Many of these incidents are linked to patient-related factors such as dementia and the inherent limitations of human monitoring, making them difficult to prevent through visual observation alone. The objective of this research is to develop a prediction and prevention system using Artificial Intelligence (AI) to enhance patient safety. The current system utilizes MediaPipe (Pose AI) to recognize body movements, enabling 24/7 patient monitoring without requiring additional labor. By analyzing actual footage, the system is designed to identify specific movement patterns that indicate a patient is dangerously close to a self-removal event. Future work involves integrating YOLO (object detection AI) to further improve recognition accuracy and system responsiveness. This AI-driven approach aims to overcome the limitations of human monitoring in dialysis settings, providing a continuous safety net for high-risk patients. Presentation Type: Pre-Recorded Presentation (https://www.youtube.com/@cwusource5518) Keywords: Dialysis Safety, AI, Media Pipe, YOLO, Body Movement Recognition SOURCE Form ID: 4N Recognition of Body Movement Using AI Tomohiro Oshiro Project Mentor(s): Mariko Oda, Kohei Arai
171
Made with FlippingBook interactive PDF creator