objdetect aruco module CORNER_REFINE_CONTOUR has no effect (regression in opencv 4.7) #23437
Description
System Information
OpenCV python version: 4.7.0.68 and 4.7.0.72
Operating System / Platform: Ubuntu 22.04
Python version: 3.10.6
Detailed description
The parameter CORNER_REFINE_CONTOUR is supposed to refine the corners, but the returned corner are still an integer value
Steps to reproduce
I have created a following code snippet to process the image
import os
import csv
import time
import datetime
import threading
import cv2
import cv2.aruco as aruco
import numpy as np
# opencv 4.7
aruco_dict = aruco.getPredefinedDictionary(dict=aruco.DICT_6X6_250)
parameters = aruco.DetectorParameters()
# opencv 4.6
#aruco_dict = aruco.Dictionary_get(dict=aruco.DICT_6X6_250)
#parameters = aruco.DetectorParameters_create()
parameters.cornerRefinementMethod = aruco.CORNER_REFINE_CONTOUR # CONTOUR works better than SUBPIX in terms of the std of the positions
parameters.cornerRefinementMaxIterations = 1000
parameters.cornerRefinementMinAccuracy = 0.001
image = cv2.imread("aruco-test-image.jpg")
markerCorners, markerIDs, rejectedImgPoints = aruco.detectMarkers(image, aruco_dict, parameters=parameters)
print(markerCorners)
In opencv 4.7 it will return without any refinement performed.
(array([[[ 361., 417.],
[1162., 741.],
[ 788., 1518.],
[ 106., 1164.]]], dtype=float32),)
This is considered a regression from opencv 4.6. Running the above code in opencv 4.6 will produce a different result.
Issue submission checklist
- I report the issue, it's not a question
- I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
- I updated to the latest OpenCV version and the issue is still there
- There is reproducer code and related data files (videos, images, onnx, etc)