Jitter in OpenCV Circle Detection

Question

I am working on a project and I need to determine the location of circles in a live feed from a camera.

I have already managed to get this kinda running, but I get a weird jitter with the recognition of the circles, and sometimes a circle is not recognized at all.

Here is a Example of this:

The function I Use for editing the Image is:

    def preprocess_image(self):
        self.image = cv2.cvtColor(self.image, cv2.COLOR_BGR2GRAY)
        
        # apply GuassianBlur to reduce noise. medianBlur is also added for smoothening, reducing noise.
        self.image = cv2.GaussianBlur(self.image,(5,5),0);
        self.image = cv2.medianBlur(self.image,5)
        
        # Adaptive Guassian Threshold is to detect sharp edges in the Image. For more information Google it.
        self.image = cv2.adaptiveThreshold(self.image,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,\
                cv2.THRESH_BINARY,11,3.5)
        
        kernel = np.ones((2,2),np.uint8)
        self.image = cv2.erode(self.image,kernel,iterations = 1) # gray = erosion
        self.image = cv2.dilate(self.image,kernel,iterations = 1) # gray = dilation    

And the Method for detecting the circles is:

    def get_circles(self):
        rows = self.image.shape[0] # width of the image (1280)
        # detect circles in the image
        self.circles = cv2.HoughCircles(
            self.image, 
            cv2.HOUGH_GRADIENT, 
            1, 
            rows / 8, 
            param1=self.parameters["param1"], 
            param2=self.parameters["param2"], 
            minRadius=self.parameters["minRadius"], 
            maxRadius=self.parameters["maxRadius"]
        )

        if self.circles is None: return None
        self.circles = np.round(self.circles[0, :]).astype("int")

How do I get rid of this jitter and the sometimes missing recognition of the circles?

Answer 1

I solved the issue by ditching hough transforms and instead using connectedComponentsWithStats to transform the circles into black blobs as suggested by Christoph Rackwitz.

Then I used the SimpleBlobDetector to detect the wanted circles.

This works way better then before!

Here is the Code I used:

import cv2
import numpy as np

def detect_circles(image):
    _, labels, _, _ = cv2.connectedComponentsWithStats(image)

    # Extract the pixels associated with the current label
    label_mask = np.zeros(image.shape, dtype=np.uint8)
    label_mask[labels == 1] = 255

    params = cv2.SimpleBlobDetector_Params()

    # Set Area filtering parameters
    params.filterByArea = True
    params.minArea = 1000
    params.maxArea = 10000

    params.filterByCircularity = True
    params.minCircularity = .7

    # Create a detector with the parameters
    detector = cv2.SimpleBlobDetector_create(params)

    # Detect blobs
    keypoints = detector.detect(label_mask)
    circles = []
    for kp in keypoints:
        circles.append(np.array([int(kp.pt[0]), int(kp.pt[1]), int(kp.size / 2)]))
    
    return circles

cap = cv2.VideoCapture("<your sample video>")

while True:
    ret, frame = cap.read()
    if np.sum(frame) == 0:[![enter image description here][1]][1]
        break

    detect_circles(frame)

    # draw the circles on image here 

    if cv2.waitKey(1) & 0xFF == ord('q'):
        break   

cv2.destroyAllWindows()
cap.release()

And this is the result: