VISIONGAUGE: A COMPUTER VISION MODEL TO DETECT AND READ U-TUBE MANOMETERS

This work proposes a computer vision model based on the sequential implementation of machine learning models for the detection and reading of U-tube manometers, named VisionGauge. The solution is composed of two stages: a detector and a regressor. The detector architecture was selected based on an evaluation among YOLOv8s, YOLO11s, and YOLO26s, while the regressor was defined through a comparative study among ResNet-18, EfficientNet-B0, MobileNetV3 Small, and MobileNetV3 Large architectures, all adapted for the regression task. Custom datasets were created for model training and for the complete evaluation of VisionGauge. The best configuration was achieved with YOLOv8s + EfficientNet-B0, reaching an F1-Score of 99.94% and an MAE of on the test set, indicating a low regression error.