Analysis of Detection and Distance Measurement of Screw Hole Using SGBM Stereo Vision Algorithm with YOLOv5

Abstract

To address the current issues of insufficient automation and low efficiency caused by manual operations in the screw tightening process at the top of transformers in factories, this paper proposes a stereo vision-based object detection and 3D ranging method utilizing YOLOv5. First, a high-precision detection model tailored for screw holes is trained using the pre-trained YOLOv5 model combined with a custom dataset collected from actual production environments. Subsequently, integrating a stereo vision system with the SGBM stereo matching algorithm enables precise 3D localization and distance measurement of target screw holes. By establishing an integrated experimental platform and implementing real-time data exchange between PLC and PC via host-to-host communication mechanisms, the feasibility and stability of this method in practical applications are validated. This approach enables automatic recognition and precise localization of screw holes, making it effectively applicable for automating upgrades in transformer assembly production lines. It significantly enhances operational efficiency and assembly quality.