Research on real-time detection algorithm for container corner fittings based on improved YOLOv8

Clearly, with the rapid advancement of global trade and logistics, corner detection plays a vital role in enhancing container loading/unloading management efficiency. Most traditional detection methods encounter the issues of applying real-time processing and achieving high robustness in various complex surroundings. The proposed method builds on the current YOLOv8 architecture, fully utilizes its advantages in real-time object detection, and makes some innovative improvements. In particular, the main contributions of this paper are as follows: (1) A multi-scale feature fusion mechanism which is proposed and adopted in the pyramid network; (2) An improved attentive module to preferentially capture the corner features produced; (3) A dynamic anchor frame adjustment strategy is proposed to adjust the sizes and orientations of the corner shapes. Using these enhancements makes the model capable of accurately and in real time detecting container corners even in low resolution image and after distorted backgrounds. From the experimental results, we can observe that the detection accuracy of the proposed algorithm is significantly better than that of the standard YOLOv8 algorithm, and the improvement of inference speed is also more obvious.