SurgeryOCR

Description

This project aims to automate the extraction and labeling of critical numerical data displayed within ophthalmological surgery videos. By leveraging CV+OCR(Optical Character Recognition) techniques, the repository provides tools to:

• Identify and extract numerical values appearing on the video feed.
• Label these extracted numbers based on the specific stage of the surgery they correspond to.
• Provide a structured dataset for further analysis of surgical procedures, potentially leading to insights for research, training, and quality improvement in ophthalmology.

Key Features

• Video Analysis: Processes ophthalmological surgery videos as input.
• On-Screen OCR: Employs OCR to recognize numerical data displayed on the video frames.
• Stage Identification: Aims to identify different stages of the surgery to provide context to the extracted data.
• Data Labeling: Labels the extracted numerical data according to the identified surgical stage
• Output: Generates structured output (e.g., jsonl) containing the extracted numerical data and their corresponding labels and timestamps.

Installation

1. Clone the repository:

   git clone <your-repository-url>
   cd SurgeryOCR

2. Install dependencies:

   pip install -r requirements.txt
   

Usage

This is the standard operating procedure for analyzing surgery videos. These commands assume you are running them from the root of the project directory.

1. Place Videos: Put all surgery videos (e.g., .mp4 files) into the data/ directory.

2. Extract ROI Images: Pre-process the videos to extract ROI images. This step significantly speeds up the subsequent analysis and UI responsiveness. The following command processes all videos in the data/ directory and extracts all configured ROIs, including saving full frames for faster UI loading.

   python extract_roi_images.py --video ./data --region all --save-full-frames

3. Run OCR Analysis: Perform change detection and OCR on the extracted ROIs. This command processes all videos and generates .jsonl files with the OCR results.

   python surgery_analysis_process.py --video ./data --region all

4. Analyze Surgery Stages: Run pattern matching to detect and segment different surgical stages based on the visual patterns in the ROIs.

   python stage_pattern_analysis.py --video ./data

5. Review Results: Launch the graphical user interface to load a video and visually inspect the analysis results, including OCR data and surgical stage segments.

   python video_annotator_gui.py