A Non-Intrusive Computer Vision Framework for Real-Time Vital Sign Digitization and Adaptive Drug Infusion in Critical Care Environments

Purpose: This work proposes a computer vision framework to automate the extraction of vital signs from bedside monitor systems and facilitate adaptive drug infusion in intensive care units (ICUs). This approach is intended to meet the requirement for less manual intervention and improved accuracy in critical care settings. Methods: An 8-megapixel camera captures time-lapse images of a simulated monitor display at a rate of 3 frames per second. Images are preprocessed (grayscale conversion, histogram stretching, edge detection, color filtering) before Tesseract OCR reads out vital signs (heart rate, blood pressure). On a synthetic dataset of 3,000 vital sign images, the system achieved 99.87% accuracy. Retrieved information inputs a pre-programmed fuzzy logic controller to adjust syringe pump infusion rates of medications like norepinephrine. The system is designed for non-invasive integration with existing monitors. Results: OCR accuracy improved from 95.43% (without preprocessing) to 99.87% on the test dataset. An optimal focal distance of 11.9 cm was determined, balancing accuracy and compactness. The fuzzy logic controller provided stepwise adjustment of flow rate based on blood pressure thresholds, with simulations using non-biological substitutes demonstrating predictable control around a 3 ml/hour baseline. Conclusion: This framework offers a promising proof-of-concept for ICU automation, with potential extensions to other monitoring areas. The primary limitation is the lack of human trials and validation in a clinical environment, implying the necessity for future clinical validation before any real-world application can be considered.