vital_extraction_from_monitors

Extracting metrics like heart rate, spo2, respiration rate, etc from images of monitors

Sample Input

After Extracting Monitor Screen and Warping

Detecting HR, SPO2, RR, BP, and HR Graph

Digitized HR Graph

Instructions for running the project

• Extract the zip file named "example_name.zip" to your desired location
• Create a folder named "*input", inside the "*inter_iit_main_folder folder", and load all the test images in this folder
• Run all the cells of the "final_submission.ipynb" notebook to load all the necessary models and other dependencies.
• The "*inference" function is present inside the "*final_submission.ipynb" notebook itself , which accepts an image path string as its only input.
• Inside "*final_submission.ipynb" run the cells under the title - "*Testing" to get all the outputs.
• The following example illustrates the output of the "inference" function: {"HR":"88", "SPO2":"98", "RR":"15", "SBP":"126"} Note: If DBP and MBP are not there in the output dictionary , they are missing
• The "create_graph_from_image" digitizes the HR graph and displays it as a matplotlib plot.

If we get "/" in our BP results, then we have split it into SBP and DBP. If not, then we have kept the output as BP.

We perform the following tasks:

1. Loading the OCR Model
2. Loading the Segmentation and monitor-layout wise classification models.
3. Loading the Yolo object detection models for respective monitor layout types.
4. The screen_from_segmentation function takes in the original image as input and returns the warped monitor screen image as output.
5. The identify_screen_type function uses the classificaiton model to classify the monitor screen image into the monitor type.
6. The create_graph_from_image function converts the graph image to digitzed output.
7. The detect_ocr function identifies characters from the objects detected by the yolo object detection model and labels them into their categories.
8. The predict function takes the yolo models and image and provides with the bounding box outputs.
9. The inference function implements the entire pipeline.