立体校准差的投影误差

如何解决立体校准差的投影误差

我执行了 OpenCV 的立体校准功能来计算两个相机之间的旋转矩阵和平移向量。然后，我打印出 RMS 误差。但是当我在另一个系统中使用相同的参数执行代码时，我得到两个不同的错误值。 我在问有两个不同的错误值的原因。

Root mean square error

for i,fname in enumerate(imagesL):
        Lframe = cv.imread(imagesL[i])
        Rframe = cv.imread(imagesR[i])
        Limg = cv.cvtColor(Lframe,cv.COLOR_BGR2GRAY)
        Rimg = cv.cvtColor(Rframe,cv.COLOR_BGR2GRAY)
        Lframe = cv.undistort(Limg,LcameraMatrix,LdistCoeffs,None,LnewCameraMatrix)
        Rframe = cv.undistort(Rimg,RcameraMatrix,RdistCoeffs,RnewCameraMatrix)
        #cv.imwrite("undistorted/L{}".format(imagesR[i][15:]),Rframe)
        #cv.imwrite("undistorted/{}.png".format(i),Lframe)
        #ImageView.frame = Rframe
        #ImageView.frame = Lframe
        retL,Lcorners = getimagePoints(Lframe,boardWidth,boardHeight)
        retR,Rcorners = getimagePoints(Rframe,boardHeight)
        if retL and retR:
            objectPoints.append(objp)
            LimagePoints.append(Lcorners)
            RimagePoints.append(Rcorners)

        #STEREO CALIBRATE
    print("calculating stereo paramaeters")
    rms,K1,D1,K2,D2,R,T,E,F = cv.stereoCalibrate(objectPoints,LimagePoints,RimagePoints,(Lframe.shape[0],Lframe.shape[1]),criteria=criteria,flags=flags)

    print("root mean square erro:\n")
    print("rmse = {}".format(rms))