[Python] Single-Image Camera Calibration Using 4+ Known Points from a Broadcast Football Match

I'm attempting to calibrate a camera from a single image captured during a broadcast football match.

My goal is to estimate the camera's intrinsic parameters using 4 known points using this function:

cv2.calibrateCamera(real_image_points, image_points, image.shape[:2], None, None)


but I'm encountering an error in OpenCV:

OpenCV(4.7.0) D:\a\opencv-python\opencv-python\opencv\modules\calib3d\src\calibration.cpp:1576: error: (-5:Bad argument) For non-planar calibration rigs, the initial intrinsic matrix must be specified in function 'cvCalibrateCamera2Internal'.
​

Is it possible to calibrate a camera from a single image captured during a broadcast football match?

Here is the full code used:

import cv2
import numpy as np



clicked_points = []


def click_event(event, x, y, flags, param):
if event == cv2.EVENT_LBUTTONDOWN:
clicked_points.append((x, y))
print(f"Clicked pixel at (x={x}, y={y})")


image=cv2.imread('image0.png')

cv2.imshow("Image", image)
cv2.setMouseCallback("Image", click_event)

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

cv2.destroyAllWindows()


print("Clicked points:")
clicked_points=np.array(clicked_points,dtype=np.float32)
for point in clicked_points:
print(point)
clicked_points=[clicked_points]
scale_x=image.shape[1]/105
scale_y=image.shape[0]/68

first_point_x=5.83*3*scale_x
second_point_x=5.83*4*scale_x
third_point_x=5.83*5*scale_x
fourth_point_x=5.83*6*scale_x
#all 4 points have y value equal to 0 in 3d world coordinate frame
distance_z_camera=77.


real_image_points=[np.array([(first_point_x,0,distance_z_camera),(second_point_x,0,distance_z_camera),(third_point_x,0,distance_z_camera),(fourth_point_x,0,distance_z_camera)],dtype=np.float32)]

retval, K, distortion_coefficients, _, _ = cv2.calibrateCamera(real_image_points, clicked_points, image.shape[:2], None, None)

Continue reading...