DEV Community

Cover image for Playing around with selenium and openCV!
Pablo Calvo for

Posted on

Playing around with selenium and openCV!

Selenium library is frequently used for UI test automation, and was probably designed for it, but selenium itself is not a testing library and people quite often forget about it

Selenium is basically a set of tools that allow your code to interact with a browser and the html elements on it.

The selenium WebDriver architecture consists in four main parts:

  • The language bindings (libraries for each language).
  • The selenium server.
  • The browser drivers.
  • Finally, the drivers.

Playing around with selenium

Selenium is not a super fast processing tool, maybe because of the many components that need to get involved. I made a fun experiment to see how slow is really selenium by reducing the external dependencies at its minimum.

For the experiment I will basically be executing clicks against a system in which no internet call is executed on each click and also there is no html rendering happening (or this does not matter that much).

Python and open CV

I used python along with openCV to open an image and generate a list of pixel points for each black pixel on the given image.

def generate_coordenates():
    print('Generating coordinates based on image')
    black_coordinates = [] #initial position

    for i in range(image.shape[0]):
        for j in range(image.shape[1]):
            # find the black pixels
            if image[i,j,0]==0 and image[i,j,1]==0 and image[i,j,2]==0: 

    print('Writing coordinates to file')
    # generate a coordinates file
    with open(COORDINATES_FILE, 'a') as out:
        for x, y in black_coordinates:

    print('Finish generating coordinates file')

Selenium and a website like MS Paint

Then I use selenium to go to a website like MSPaint ( and generate an image based on the black pixels.

driver = webdriver.Chrome(DRIVER_LOCATION)
        with open(COORDINATES_FILE) as file:

            # move to starting point
            actions = ActionChains(driver)
            actions.move_by_offset(STARTING_COORDINATES.get('x'), STARTING_COORDINATES.get('y'))
            # start painting
            for line in map(lambda line: line.rstrip('\n'), file):
                print('Creating move actions')
                xy = line.split(',') # split the x and y
                current = {'x': int(xy[0]), 'y': int(xy[1])} 
                move_to = {'x': current.get('x') - before.get('x'), 
                           'y': current.get('y') - before.get('y')}

                actions.move_by_offset(move_to.get('x'), move_to.get('y'))
                before = current.copy()

            # finish painting

The process

A couple of minutes:
a couple of minutes

20 minutes:
15 minutes

50 minutes:
50 minutes

The results

The coordinates file ended up with 3777 rows. Because for each row selenium executes 2 actions: move_by_offset and click it means 7554 selenium actions were be executed.

The test took 54.67 minute to complete which means around 3280 seconds.

And if we do the math: 7554 / 3280 = ~2.30. It means that 2.30 selenium actions were executed per second.

Quite fun...


cheers :) and remember to keep using selenium

if you liked the post:

Buy Me A Coffee

Top comments (0)