如何使用 OpenCV 通过solvePnP 估计距离? [英] How to estimate distance by solvePnP using OpenCV?
本文介绍了如何使用 OpenCV 通过solvePnP 估计距离?的处理方法，对大家解决问题具有一定的参考价值，需要的朋友们下面随着小编来一起学习吧！
问题描述
我尝试估计相机和用于校准的棋盘之间的距离.但是我在棋盘上的点的 Z 值(993 - 1021 毫米)上有很大的不同.棋盘平行于相机，所以我预计相机中心和棋盘平面上任何点之间的 Z 值应该是相同的.
I tried to estimate distance between my camera and a chessboard which I used for the calibration. But I've got a big different in the Z-values (993 - 1021 mm) for points on the chessboard. The chessboard is parallel to the camera, so I expected that Z-values between the camera centre and any point on the chessboard plane should be the same.
我做错了什么?
代码:
import numpy
import cv2
#from cv2.calibrateCamera.
imagePoints = numpy.array([[185.38619995, 88.96965027],
[215.52053833, 89.59921265],
[245.42134094, 90.34884644],
[275.23474121, 90.67272186],
[304.63644409, 91.27327728],
[333.90859985, 91.75786591],
[185.34030151, 118.99983978],
[215.52850342, 119.46482849],
[245.514328, 119.89020538],
[275.03085327, 120.42667389],
[304.64593506, 120.67990112],
[333.73968506, 121.14579773],
[185.34428406, 148.93850708],
[215.58241272, 149.32409668],
[245.4022522, 149.86160278],
[275.17852783, 149.81486511],
[304.5428772, 150.28009033],
[333.79421997, 150.35099792],
[185.18565369, 178.83206177],
[215.445755, 179.09082031],
[245.40261841, 179.29736328],
[275.07699585, 179.61761475],
[304.49923706, 179.56680298],
[333.78756714, 179.81796265],
[185.00271606, 208.8531189],
[215.40895081, 208.91560364],
[245.23666382, 209.10614014],
[275.06970215, 209.07598877],
[304.59561157, 209.13166809],
[333.90438843, 209.09313965],
[184.86560059, 238.93251038],
[215.23825073, 238.76199341],
[245.25619507, 238.73907471],
[274.97564697, 238.73266602],
[304.59933472, 238.72055054],
[333.96078491, 238.68598938],
[184.71801758, 269.20956421],
[215.05563354, 268.91809082],
[245.1418457, 268.70123291],
[274.9392395, 268.59295654],
[304.65649414, 268.48577881],
[333.98513794, 268.29562378],
[184.43510437, 299.72283936],
[214.71133423, 299.52212524],
[244.97724915, 298.88729858],
[275.03652954, 298.61529541],
[304.74026489, 298.40454102],
[334.23959351, 298.21008301]])
objectPoints = numpy.array([[0., 0., 0.],
[0.083, 0., 0.],
[0.16599999, 0., 0.],
[0.24899998, 0., 0.],
[0.33199999, 0., 0.],
[0.41499999, 0., 0.],
[0., 0.083, 0.],
[0.083, 0.083, 0.],
[0.16599999, 0.083, 0.],
[0.24899998, 0.083, 0.],
[0.33199999, 0.083, 0.],
[0.41499999, 0.083, 0.],
[0., 0.16599999, 0.],
[0.083, 0.16599999, 0.],
[0.16599999, 0.16599999, 0.],
[0.24899998, 0.16599999, 0.],
[0.33199999, 0.16599999, 0.],
[0.41499999, 0.16599999, 0.],
[0., 0.24899998, 0.],
[0.083, 0.24899998, 0.],
[0.16599999, 0.24899998, 0.],
[0.24899998, 0.24899998, 0.],
[0.33199999, 0.24899998, 0.],
[0.41499999, 0.24899998, 0.],
[0., 0.33199999, 0.],
[0.083, 0.33199999, 0.],
[0.16599999, 0.33199999, 0.],
[0.24899998, 0.33199999, 0.],
[0.33199999, 0.33199999, 0.],
[0.41499999, 0.33199999, 0.],
[0., 0.41499999, 0.],
[0.083, 0.41499999, 0.],
[0.16599999, 0.41499999, 0.],
[0.24899998, 0.41499999, 0.],
[0.33199999, 0.41499999, 0.],
[0.41499999, 0.41499999, 0.],
[0., 0.49799997, 0.],
[0.083, 0.49799997, 0.],
[0.16599999, 0.49799997, 0.],
[0.24899998, 0.49799997, 0.],
[0.33199999, 0.49799997, 0.],
[0.41499999, 0.49799997, 0.],
[0., 0.58099997, 0.],
[0.083, 0.58099997, 0.],
[0.16599999, 0.58099997, 0.],
[0.24899998, 0.58099997, 0.],
[0.33199999, 0.58099997, 0.],
[0.41499999, 0.58099997, 0.]])
cameraMatrix = numpy.array([[364.21551784, 0., 253.82602981],
[0., 363.51882052, 206.4493302],
[0., 0., 1.]])
distortionCoefficient = numpy.array([[0.09360905, -0.37480762, 0.00147107, -0.00132464, 0.25238534]])
Z = numpy.zeros((48))
retval,R, T = cv2.solvePnP(objectPoints,imagePoints,cameraMatrix,distortionCoefficient) #from chessboard coordinate space to camera coordinate space
R,jacobian = cv2.Rodrigues(R) #from R-vector to R-matrix
for i in range(0,48):
point = numpy.dot(objectPoints[i],R) + numpy.matrix.transpose(T)
Z[i] = point[0,2] * 1000 #Z-value to mm
print(Z.max() - Z.min())
print(Z)
棋盘图像
推荐答案
好吧，这似乎是我的错误.可能电路板与相机并没有真正平行，尽管我试图把它弄对.如果我们查看计算出的 Z 坐标，我们可以看到 Z 坐标从右下角到左上角线性减小.
Well, it seems to be my mistake. Probably the board was not really parallel to the camera, although I tried to put it right. If we look to computed Z coordinates, we can see that the Z coordinate decreases linearly from bottom right corner to top left corner.
print(numpy.round(Z.reshape(8,6)))
[[ 1013. 1009. 1005. 1001. 997. 993.]
[ 1014. 1010. 1006. 1002. 998. 994.]
[ 1015. 1011. 1007. 1003. 1000. 996.]
[ 1016. 1012. 1009. 1005. 1001. 997.]
[ 1018. 1014. 1010. 1006. 1002. 998.]
[ 1019. 1015. 1011. 1007. 1003. 999.]
[ 1020. 1016. 1012. 1008. 1004. 1001.]
[ 1021. 1017. 1013. 1010. 1006. 1002.]]
这篇关于如何使用 OpenCV 通过solvePnP 估计距离?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
查看全文
