Number recognizer

It is number recognizer on videos. You can configurate video preprocessing, vision window size and check results after recognizing on flow. This project in active working and I use it to get lab research results without having to look at the indicators every minute)

It uses easyOCR package to recognize. For details and installation instructions see: https://github.com/JaidedAI/EasyOCR. Image processing packadge is OpenCV.

Installation

Please install it in this order, because openCV can be crashed. For GPU:

• Install CUDA for working on GPU: https://developer.nvidia.com/cuda-toolkit-archive
• Install cuDNN SDK (I don't know it nessesary, but i install it for using GPU ): https://developer.nvidia.com/cudnn

For CPU and GPU:

• Install Tesseract OCR engine (necessary for easyOCR package) : https://tesseract-ocr.github.io/tessdoc/Installation.html After installation add TesseractOCR folder to PATH (in windows)

• Install pythorch: https://pytorch.org/get-started/locally/

• Install openCV for python: https://opencv.org/get-started/

• Install easyOCR: https://github.com/JaidedAI/EasyOCR

  After installation easyOCR restart your computer and after delete opencv-python-headless package to fix crashing with MethodNotImplemented error from cv2.selectROI function (it provides image cutting interactively), because opencv-python-headless not implement GUI functions and used for servers:

  pip uninstall opencv-python-headless I can't fix it yet, and can't understand this problem, but several combinations this actions may help to install

• Install python packadges for visualisation and animation:

  • matplotlib
  • PyQt5(for animation in separate window)
  • pympl (for animation in jupyter notebook)
  • tqdm (for animate recognize process) Example requirements file in repository (for gpu and cpu)

Features:

• Process every video frame to increase recognizing quality interactively
• Choose framerate you need to recognize
• Cut several areas to recognize numbers on playing video
• Configure custom pattern using regexp to check the correctness of recognizing and use slightly wrong results to combine them to get fully correct with verbose
• Configure smart searching of image preprocessing configurations in case uncorrect recognizing
• Easy to configure and use

Plans:

• Add contrast and more options to image processor
• Update image processor sweep to find best conditions faster
• Add value corrector which uses adjacent frames
• Add timer and zoom on video
• Contain it to docker image