Gas Pump OCR

Project for attempting to scan pictures of gas pumps and detect the digits in the cost and fuel amount displays. Operating under the assumption that most pumps use a 7 segment digit display.

Dependencies

• Python
• OpenCV
• NumPy

Global Setup

• brew install python - Using python 2.7.13
• brew install opencv3 - Using >= OpenCV 3.2.0
• Install the OpenCV package into the python site packages following the instructions from the above command output

Virtual Env

• pip install virtualenv - Install virtualenv if you haven't
• virtualenv GasPumpOCR_Env - Create virtual env
• source GasPumpOCR_Env/bin/activate - Activate virtual env
• pip install -r requirements.txt - Install dependencies
• Write/run the code
• deactivate - When you are done and want the global python env

virtualenvwrapper

virtualenvwrapper is a nice set of wrappers around virtualenv that make it easier to use

Install and configure virtualenvwrapper

pip install virtualenvwrapper
export WORKON_HOME=$HOME/.virtualenvs
source /usr/local/bin/virtualenvwrapper.sh

Create new env

mkvirtualenv GasPumpOCR_Env

Work on a virtual env

workon GasPumpOCR_Env

Install dependencies

pip install -r requirements.txt

Deactivate

deactivate

Running

• playground.py [file_name] - Used to try out different image manipulation variables, can hardcode an image or pass one in via the command line
• train_model.py - Used to take a folder organization of confirmed digits and generate a knn training file
• test_processing.py - Used to run the trained algorithm on a folder of test images to test accuracy, can also be setup to test all the configurations to determine optimal values
• generate_distorted_images.py - Used to take an image or a folder and run a set of image manipulations to create addtional sample images

Screenshot