Research on a Dual-View Object Detection Algori thm Based on YOLOv5n and SORT

With the increasing demand for safety monitoring in the metallurgy industry, intelligent monitoring systems have made significant progress. However, the accuracy of object detection faces numerous challenges due to smelting workers often facing away from cameras and frequent on-site occlusions. To address this issue, this paper proposes a dual-view object recognition method based on YOLOv5n and the SORT algorithm. In this method, we first optimize the YOLOv5n model by incorporating an EMA attention mechanism to reduce occlusions and background interference. Additionally, to enhance the detection accuracy of small objects, we add a P2 detection head to the model, significantly improving recall and precision. Experimental results show that PE-YOLO improves detection accuracy by 4.2% compared to the baseline model, with enhancements in both precision and recall. Secondly, this paper improves the SORT algorithm by introducing dual-view information to optimize the object matching strategy. By employing coordinate mapping techniques, we effectively reduce viewpoint occlusions and mismatches, ensuring accurate object matching under two viewpoints. Experimental data indicate that the recognition rate of safety equipment increases from 22.1% under a single viewpoint to 49.5% under dual viewpoints, demonstrating a significant improvement. In summary, the dual-view object recognition method proposed in this paper successfully addresses the challenge of safety equipment recognition in the metallurgy industry and has broad application prospects in industrial safety monitoring.