count red color object from video opencv python

Here's how I made it: first, subtract blue and green channels from the red channel to find the "red" pixels. The result looks like this:

Threshold works really well on this, obviously:

From here we can find contours:

Count them, and that's the number of stripes. But there's also connectedComponents function, it can count those without finding contours.

import cv2
import numpy as np

#load a frame as int16 array to avoid overflow etc. when subtracting
img = cv2.imread("test_images/redbars.png").astype(np.int16)

#separating into blue, green and red channels
img_b = img[:, :, 0]
img_g = img[:, :, 1]
img_r = img[:, :, 2]

#for a red stripe, the red component should be much bigger than the rest
res_br = img_r - img_b
res_gr = img_r - img_g
res = np.maximum(res_br, res_gr)
res[res < 0] = 0
res = res.astype(np.uint8)  #convert back to uint8 for findContours etc.

ret, thresh = cv2.threshold(res, 50, 1, cv2.THRESH_BINARY)

#find contours and count them
im2, contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

#gives the correct output of 5
print(len(contours))

#alternatively, count connected components with this:
ncomp, nimg = cv2.connectedComponents(thresh)
print(ncomp - 1)  #it counts the background as one of the components, so subtract 1

Here is a pipeline you can use to count the number of red strips on a frame:

int main(int argc, char** argv)
{
    Mat input = imread("red_lines.png");
    Mat hsv;
    cvtColor(input, hsv, CV_BGR2HSV);

    Mat mask;
    inRange(hsv, Scalar(0, 0,0), Scalar(15, 255,255), mask);

    Mat kernel = getStructuringElement(MORPH_ELLIPSE, Size(7, 7));
    erode(mask, mask, kernel);
    dilate(mask, mask, kernel);

    vector<vector<Point>> contours;
    findContours(mask, contours, RETR_EXTERNAL, CHAIN_APPROX_NONE);
    cout << "objects found: " << contours.size() << endl;

    for (int i = 0; i < contours.size(); i++)
    {
        drawContours(input, contours, i, Scalar(0, 255, 0), 2);
    }

    imshow("mask", mask);
    imshow("contours", input);
    waitKey(0);

    return 0;
}

The important part in my code, is the morphological operations (erosion then dilation) I apply on the mask (resulting from inRange). This allows to remove the "mask's noise" and thus to catch only the big red strips.

To further improve the robustness, I recommend you to check the area of each contour. If it is above a certain threshold, just discard it.

P.S. This is C++ code, but I'm sure you'll figure out how to translate it into Python code... :)