Infusion Monitoring and Real-Time Prediction Based on YOLOv8 with P2 Enhancement

In the context of smart healthcare, infusion monitoring systems are critical for ensuring patient safety and preventing infusion interruptions. This paper addresses the challenge of insufficient small-target detection accuracy in long-distance infusion monitoring by proposing a non-invasive, computer vision-based solution. In the proposed scheme, a P2 layer is integrated into the YOLOv8 model to enhance shallow feature retention and improve target boundary regression. A novel two-stage detection strategy is proposed, where the infusion bottle is first localized in the full image, and subsequently, the liquid level is precisely detected from a cropped region, effectively reducing background interference and enhancing localization accuracy. Furthermore, the system incorporates DeepSORT tracking combined with a sliding-window regression approach to dynamically estimate the infusion rate and predict the remaining infusion time in real time. Experimental results demonstrate that the proposed method significantly improves small-target detection and infusion prediction accuracy, offering robust technical support for automated infusion management in smart healthcare.