Intravenous (IV) Therapy Infusion Drip Image

We introduce a high-performance computer vision based Intraveneous (IV) infusion speed measurement system as a camera application on an iPhone or Android phone. Our system uses You Only Look Once version 5 (YOLOv5) as it was designed for real-time object detection, making it substantially faster than two-stage algorithms such as R-CNN. In addition, YOLOv5 offers greater precision than its predecessors, making it more competitive with other object detection methods. However, YOLOv5 can be challenging to use on a mobile device for several reasons as it requires substantial computational resources for image processing and prediction generation. Thus, we chose the model optimization approach because it requires the least effort to implement. Because NCNN (Neural Network Computing) is a high-performance neural network inference framework optimized for mobile platforms such as Android and iOS, we converted a YOLOv5 model to an NCNN (Novel Convolutional Neural Network) model. Compared to the previous research, our application showed less variability and higher consistency in the infusion flow rate measurement.

本团队推出了一套基于高性能计算机视觉的静脉注射（IV）输液速度测量系统，该系统作为一款适用于iPhone或Android手机的相机应用程序。本系统采用You Only Look Once版本5（YOLOv5）作为其基础，因其专为实时目标检测而设计，相较于R-CNN等两阶段算法，具有显著的速度优势。此外，YOLOv5相较于其前辈模型，在精度方面亦有所提升，从而在与其他目标检测方法竞争中更具竞争力。然而，由于YOLOv5在移动设备上的应用存在一定挑战，主要原因是其图像处理和预测生成过程需要大量的计算资源，因此我们选择了模型优化方法，因其实施成本相对较低。鉴于NCNN（Neural Network Computing）是一个针对Android和iOS等移动平台进行优化的高性能神经网络推理框架，我们将YOLOv5模型转换为NCNN（Novel Convolutional Neural Network）模型。与先前研究相比，我们的应用在输液速度测量方面表现出了更小的变异性及更高的一致性。