0
$\begingroup$

I am trying to detect meteors in a video, and so far, what I did.

  1. Preprocessing: changing to grayscale and morphological operations like dilation and erosion.
  2. background subtraction method
  3. adding images to see a pattern or event in the video.

What i got is, output But now how i say that meteor is detected? Due to higher sampling rate, i get a dashed segment rather than continuous line. I tried changing the iteration values for erosion and dilation but that amplified the noises in the background too.

import cv2
import numpy as np

def preprocessing(frame):
    # Apply the preprocessing steps to a single frame
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    blur = cv2.GaussianBlur(gray, (5, 5), 0)
    #background subtraction method 
    fgmask = backSub.apply(blur)
    #apply morphological operations
    eroded_mask = cv2.erode(fgmask, kernel, iterations=2)
    dilated_mask = cv2.dilate(eroded_mask, kernel, iterations=5)
    return dilated_mask

def add(new_, frame_, i):
    #add 20 frames and return the combined image
    if(i<=20):
        new_ += frame_
        i += 1
    else:
        i = 0
        new_ = np.zeros_like(frame_)
    return new_, i
        
def meteor_detection(new_, count):
    imgLines= cv2.HoughLinesP(new_,15,np.pi/180,10, minLineLength = 400, maxLineGap = 50)
    if(imgLines is not None):
        for i in range(len(imgLines)):
            for x1,y1,x2,y2 in imgLines[i]:
                dist = ((x2-x1)**2 + (y2 - y1)**2)**(1/2)
                if(dist < 1000):
                    print(dist, count)
#                     cv2.line(new_,(x1,y1),(x2,y2),(255,255,255),2)
    return count
       
# Create background subtractor and kernel
backSub = cv2.createBackgroundSubtractorMOG2(history=5, varThreshold=10)
kernel_size = 3
kernel = np.ones((kernel_size, kernel_size), np.uint8)

# Open video capture (change 0 to the appropriate video source or file path)
video_path = path...
i = 0 #counter for images to be combined
count = 0  #a general counter for frames
cap = cv2.VideoCapture(video_path)
ret, frame = cap.read()
new_ = np.zeros_like(cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY))
while True:
    
    # Read a frame from the video stream
    ret, frame = cap.read()
    
    
    if not ret:
        break

    # Preprocess the frame
    processed_frame = preprocessing(frame)
    #add frames
    new_, i = add(new_, processed_frame, i)
    #after each 20 frames, check on the combined frames to see if there is any line, 
    if(i == 20):
        count = meteor_detection(new_, count)

    # Display the processed frame
    cv2.namedWindow("output")  
    resized_image = cv2.resize(new_, (600, 600))
    cv2.putText(resized_image, str(count), (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2)

    cv2.imshow('output', resized_image)
    count += 1
    
    # Break the loop if 'q' is pressed
    if cv2.waitKey(10) & 0xFF == ord('q'):
        break

# Release the video capture and close the windows
cap.release()
cv2.destroyAllWindows()

with hough's transform, i got some glitches and false errors.

$\endgroup$
2
  • $\begingroup$ It looks like there's three "dashes" with two spaces. Is this always the case, or if not, is the number of "dashes" always known? Is the length of the line known based on shutter speed, or are there differences in speed and/or distance that makes this not the case. Please edit your question with this information. $\endgroup$
    – TimWescott
    May 21, 2023 at 2:47
  • $\begingroup$ You say "with hough's transform, I got some glitches and false errors". Do you mean you got some false positives? Again -- please edit your question if you can clarify. $\endgroup$
    – TimWescott
    May 21, 2023 at 3:04

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.