[Python] Can someone help me understand why my corners detection code is not working?

I want to do camera calibration using a chessboard. And to do that i have developped a code based on some corner detection scripts I found on internet.

The actual issue i am facing is that the my script is able to detect corners on a chessboard whose image has been taken by a camera (let's call it camera 1). But the same script is unable to dectect corners on other chessboards whose images has been taken with two other cameras (let's say camera 2 and camera 3).

I've tried many things to understand and solve the issue, but all my attempts have failed. For eg , I've tried the script on the same image while varying the my specified chessboard size but this attempt failed. I've also tried to take images a different period of the day (because I was wondering if ambient luminosity affect the detection) but with my failed tests I guess the issue was not there.

Currently, I am lost. I don't what else I can try.

I attach to this description one image of chessboard from each camera.camera1_chessboard, camera2_chessboard, camera3_chessboard

[camera1_chessboard][1]
[camera2_chessboard][2]
[camera3_chessboard][3]


Here is my code:

import numpy as np
import cv2 as cv
import glob
import pickle
import os

chessboardSize = (50,15) # Nombre de coins dans chaque direction du damier

# Critère d'arrêt des itérations de recherches de coins
# 30 itérations max ou erreur entre 2 itérations < 0.001
criteria = (cv.TERM_CRITERIA_EPS + cv.TERM_CRITERIA_MAX_ITER, 30, 0.001)

threedpoints = [] # Vecteur de points 3D
twodpoints = [] # Vecteur de points 2D

objectp3d = np.zeros((1, chessboardSize[0]*chessboardSize[1], 3), np.float32)
objectp3d[0, :, :2] = np.mgrid[:chessboardSize[0], :chessboardSize[1]].T.reshape(-1,
2)


path = "Camera2.jpeg"

image = cv.imread(path)
img_gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)

# Si nombre de coin attendu est trouvé alors return cornersFound = True
cornersFound, corners = cv.findChessboardCorners(img_gray, chessboardSize, None)
# If found, add object points, image points (after refining them)
print(cornersFound)
if cornersFound == True:
threedpoints.append(objectp3d)
corners2 = cv.cornerSubPix(img_gray,corners, (11,11), (-1,-1), criteria)
twodpoints.append(corners2)
# Draw and display the corners
for corner in corners2:
x, y = corner.ravel()
cv.circle(image, (int(x), int(y)), 4, (0, 0, 255), -1)

####################### Nomenclature de l'image avec coins détectés
#########################################################
directory, file = os.path.split(path)
base_name, _ = os.path.splitext(file)

new_format = ".jpg"
coin_name = f"{base_name}_Corners{new_format}"
cv.imshow('img', image)
cv.waitKey(0)
cv.imwrite(coin_name, image)
cv.destroyAllWindows()

ret, cameraMatrix, dist, rvecs, tvecs = cv.calibrateCamera(threedpoints, twodpoints,
img_gray.shape[::-1], None, None)

print("\nCamera Calibrated: ", ret) # ret est l'erreur de calibration
print("\nCamera Matrix:\n", cameraMatrix)
print("\nDistorsion Coefficients:\n", dist)
# print("\nRotation Vectors:\n", rvecs)
# print("\nTranslation Vectors:\n", tvecs)


Can someone help me? please!

Continue reading...