我注意到
cv::drawFrameAxes()
功能存在问题。当三重线到达图像之外很远时,这些线就会扭曲,不再与它们应该出现的轴重合。
我正在使用 OpenCV+Contrib 4.10.0 检测 charuco 板。一旦我检测并估计了它的姿态,我就会尝试绘制轴;但是,如果我改变它们的长度,即使一切保持不变,我也可以看到不同的轴。我提供了示例图像、相机参数和要重现的代码。另请查看随附的图像。轴甚至会“断开”,或者仅保留轮廓而不填充,轴长度较长,例如10.
如果我在模拟图像以及 OpenCV (opencv-python) 4.7.0.72 上执行此操作,我会得到相同的结果。
import os
import numpy as np
import matplotlib.pyplot as plt
import cv2
# camera params after calibration
mtx = np.array([[3341.0930071 , 0. , 2020.04757797],
[ 0. , 3332.54238818, 1108.10670945],
[ 0. , 0. , 1. ]])
dist = np.array([-0.10474984, 0.04565679, 0.00948721, 0. , 0. ])
img_rgb = cv2.imread("some img")
img_rgb = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2BGR)
img = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2GRAY)
# board info
dictionary = cv2.aruco.getPredefinedDictionary(cv2.aruco.DICT_4X4_100)
board = cv2.aruco.CharucoBoard((7, 5), 0.055, 0.041, dictionary)
params = cv2.aruco.DetectorParameters()
detector = cv2.aruco.ArucoDetector(dictionary, params)
# detect markers and charuco corners
marker_corners, marker_ids, rejectedCandidates = detector.detectMarkers(img)
ret, charucoCorners, charucoIds = cv2.aruco.interpolateCornersCharuco(marker_corners, marker_ids, img, board)
ret, rvec, tvec = cv2.aruco.estimatePoseCharucoBoard(charucoCorners, charucoIds, board, mtx, dist, None, None)
# start plotting
axis_size = 0.1
result_short = cv2.drawFrameAxes(np.copy(img_rgb), mtx, dist, rvec, tvec, axis_size)
axis_size = 5
result_long = cv2.drawFrameAxes(np.copy(img_rgb), mtx, dist, rvec, tvec, axis_size)
# plot
plt.subplot(1,2,1)
plt.imshow(result_short)
plt.subplot(1,2,2)
plt.imshow(result_long)
plt.show()
result_both = cv2.drawFrameAxes(np.copy(img_rgb), mtx, dist, rvec, tvec, 0.2)
result_both = cv2.drawFrameAxes(result_both, mtx, dist, rvec, tvec, 2)
编辑:将示例图片更改为未裁剪的版本,将变量名称 mtx2 更改为 mtx,添加更多相关信息。
这已被其他人复制并已报告为 OpenCV GitHub 上的错误。