Hyper Speed OCR for medical applications

This project aims to develop a high-speed Optical Character Recognition (OCR) system tailored for ICU monitor screens, focusing on rapid and accurate recognition of digits. The objective is to enable real-time monitoring of vital signs, enhancing patient care in critical care settings. The speed of the final model 300 micro seconds with an accuracy of 98.6%

Datasets

Project consists of 4 main datasets

Contours

This contains contours corresponding to each digit

core.v8i.tensorflow

This is ther original Roboflow Data set version which we preprocessed

Number_Dataset

This is an unlabbeled Dataset which was made by crpping and preprocessing core.v8i.tensorflow

dataset_of_screen

This is a labelled dataset classified using a slow yet highly accurate OCR after seperating the digits from Number_Dataset

moments.csv

This File contains moments of the Shape of the Digits

labels.csv

This File contains labels of the Shape of the Digits

Training/Exploratory Notebooks

Preprocess.ipynb

This Notebook was used for exploring various types of preprocessing and feature extraction for this purpose

Train.ipynb

This is the training script for the Random Forest and Hue Shape Matching Algorithms

Utils

models.pkl

Final Weights of the Random Forest Classifier

app.py

Final user interface for the project

How to run app.py

Requires python 3.10 or above

python -m pip install requirements.txt

python app.py

Training script

Train.ipynb can be used to modify the training script