Automatic Detection and Counting of External Thermal Insulation Anchors Using YOLOv12n-Based Tiled Image Processing
9th International Project and Construction Management Conference (IPCMC2026), Gaziantep, Türkiye, 16 - 18 Ekim 2026, ss.1333-1342, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Basıldığı Şehir: Gaziantep
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.1333-1342
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
The adequate and proper installation of external thermal insulation anchors is important for system safety and application quality. In conventional field inspections, manually identifying and counting anchors is time-consuming and prone to user-related errors. In this study, a YOLOv12n-based object detection approach was developed for the automatic detection and counting of external thermal insulation anchors from real field images. To improve the detection of small-scale anchors in large facade images, the images were divided into overlapping tiles, and tile-based detections were subsequently mapped back to the original image coordinates. Duplicate detections in overlapping regions were eliminated using global non-maximum suppression (NMS). Model performance was evaluated using object detection and image-based counting metrics. The results showed meaningful performance in automatically identifying and counting anchors under real field conditions, with a strong linear relationship between actual and predicted anchor counts. However, the remaining detection and counting errors indicate that further performance improvements are required before direct application in field inspection. Overall, combining YOLOv12n with overlapping image tiling provides a promising approach for the automatic detection and counting of small-scale external thermal insulation anchors in real field images.