I am dealing with the estimated distance of an object with the solvePnP method. I already did some measurements and my calculated distance is far above the actual distance. I also used a calibrated camera. For the distance calculation, I use the result distance in 3 axes from solvePnP and distance pattern in 3D:
tvec = cv2.solvePnP(
objectPoints = model,
imagePoints = key_points,
cameraMatrix = camera_matrix,
distCoeffs = dist_coeffs,
flags=cv2.SOLVEPNP_ITERATIVE
)[2]
pnp_distance = mt.sqrt(tvec[0]**2 + tvec[1]**2 + tvec[2]**2)
On the image below you can see that calculated measurement (dots) are above real values (line):
Are there some extra options to make solvePnP more precise, or maybe I did something wrong?
In the beginning I used the calculated camera matrix from image size. With that and plotted error distribution I decided to divide x axis distance by 2 and that actually helped. Calculated values were much closer than with the actual camera matrix.
How I can improve my distance estimation with solvePnP and camera matrix? Thanks in advance!
