OpenCV Python: Reading and setting every pixel too slow

Question

I'm trying to track motion with OpenCV in Python. If a pixel doesn't match the color of the last frame then it gets set to black, otherwise it is set to white if it's static. I got this working pretty good, but if I try to do this with every pixel one by one, performance takes a big hit and it runs too slow.

My code:

import numpy as np
import cv2

def distance(b1, g1, r1, b2, g2, r2):
    return abs(b2 - b1) + abs(g2 - g1) + abs(r2 - r1)

pixelStep = 1
lastFrame = None
thresh = 100
cap = cv2.VideoCapture(0)

while(True):
    flag, frame = cap.read()
    frameInst = frame.copy()
    height = np.size(frame, 0)
    width = np.size(frame, 1)

    if lastFrame != None:
        for x in range(0, height, pixelStep):
            for y in range(0, width, pixelStep):
                b1 = lastFrame.item(x, y, 0)
                g1 = lastFrame.item(x, y, 1)
                r1 = lastFrame.item(x, y, 2)
                b2 = frame.item(x, y, 0)
                g2 = frame.item(x, y, 1)
                r2 = frame.item(x, y, 2)            
                dist = distance(b2, g2, r2, b1, g1, r1)
                colorValue = 255
                if dist > thresh:
                    colorValue = 0 # Change to black if there's another change from pixel
                frame.itemset(x, y, 0, colorValue)
                frame.itemset(x, y, 1, colorValue)
                frame.itemset(x, y, 2, colorValue)

    lastFrame = frameInst
    cv2.imshow('frame', frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

If I change pixelStep to 3 it'll run fast and feel right. Am I doing this right or do I need to approach this in a different way?

Just a note I originally had the resolution at 640x480. After scaling it down I can read and set all pixels and it works fast. Is this the strategy I need to go for? — Green Cell
– Green Cell, Commented Apr 8, 2015 at 15:22

theamk · Accepted Answer · 2015-04-08 16:00:57Z

2

As a rule of thumb, any python program which works on every pixel will be too slow. Opencv has huge number of functions -- use them! In your case, you can use chain four functions:

'absdiff' (http://docs.opencv.org/modules/core/doc/operations_on_arrays.html#absdiff) to get image with differences only
'split' to get three separate arrays for R, G, B differences
'add' to add them all together
'Threshold' (http://docs.opencv.org/modules/imgproc/doc/miscellaneous_transformations.html#threshold) to make a single array which shows differences

If you then want to find the areas where differences are, you can use blob detector to give you a list of areas.

answered Apr 8, 2015 at 16:00

theamk

1,6878 silver badges17 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

Mark Ransom Over a year ago

It's not just Python, it's true for any language. The overhead of making function calls to get and set a pixel value are often of the same order or greater than the operation you want to perform on the pixel.

Green Cell Over a year ago

@theamk Thanks! I wasn't aware of these kind of functions. I rewrote it and the performance is much better now.

theamk Over a year ago

@MarkRansom: well, C/C++ with proper array access (via macros/inlined functions) can be fast enough -- this is how OpenCV is implemented on generic platforms

Collectives™ on Stack Overflow

OpenCV Python: Reading and setting every pixel too slow

1 Answer 1

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related