opencv 相机标定学习记录

这两天学习相机标定，根据opencv的文档（OpenCV: Camera Calibration），整理的代码

包含了角点检测、标定、remap、重投影误差计算。在小的标定板上，findChessboardCorners和cornerSubPix的差异很小。还要再验证下大的标定板。

#-*-coding:utf-8-*-

from audioop import error
import numpy as np
import cv2
import os
import copy

path = 'D:\\try\\calib_img'
save = 'D:\\try\\calib_img\\result'
save2 = 'D:\\try\\calib_img\\result2'
calib_path = 'D:\\try\\calib_img\\calibresult'
remap_path = 'D:\\try\\calib_img\\remap'

grays = []
names = []
pObjs = []
pImgs = []

w, h = 9, 6 #corners per a chessboard row and column
pobj = np.zeros((w*h, 3), np.float32)
pobj[:, :2] = np.mgrid[0:w, 0:h].T.reshape(-1, 2)
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 0.01)

f = os.listdir(path)
'''
step one: find chessboard corners and drawchessboardCorners
'''
for i in f:
    fname = os.path.join(path, i)
    if(os.path.isdir(fname)):
        continue
    img = cv2.imread(fname)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    ret, corners = cv2.findChessboardCorners(gray, (w, h), None)

    if(ret == True):
        print(i, '-----find chessboard cornors ')
        grays.append(gray)
        names.append(i)

        corners2 = cv2.cornerSubPix(gray, corners.astype(np.float32), (5, 5), (-1, -1), criteria)
        pImgs.append(corners2)
        pObjs.append(pobj)

        img2 = copy.deepcopy(img)

        cv2.drawChessboardCorners(img, (w ,h), corners, ret)
        save_name = os.path.join(save, i)
        cv2.imwrite(save_name, img)

        cv2.drawChessboardCorners(img2, (w, h), corners, ret)
        save_name2 = os.path.join(save2, i)
        cv2.imwrite(save_name2, img2)
    else:
        continue
'''
step 2: calibrate
'''
ret, matrix, dist, r_vecs, t_vecs = cv2.calibrateCamera(pObjs, pImgs, grays[0].shape[::-1], None, None)
'''
step 3: undistor and remap
'''
for i in range(len(grays)):
    img_h, img_w = grays[i].shape[::-1]
    newMtx, roi = cv2.getOptimalNewCameraMatrix(matrix, dist, (img_w, img_h), 1, (img_w, img_h))
    x, y, w1, h1 = roi
    #undistort
    dst = cv2.undistort(grays[i], matrix, dist, None, newMtx)
    dst = dst[y:y+h1, x:w+w1]
    _name = os.path.join(calib_path, names[i])
    cv2.imwrite(_name, dst)
    #remapping
    mapx, mapy = cv2.initUndistortRectifyMap(matrix, dist, None, newMtx, (img_w, img_h), 5)
    dst2 = cv2.remap(grays[i], mapx, mapy, cv2.INTER_LINEAR)
    dst2 = dst2[y:y+h1, x:x+w1]
    _remap = os.path.join(remap_path, names[i])      
    cv2.imwrite(_remap, dst2)
'''
step 4: calculate means
'''
mean_error = 0
for i in range(len(pObjs)):
    img2_obj, _ = cv2.projectPoints(pObjs[i], r_vecs[i], t_vecs[i], matrix, dist)
    ierror = cv2.norm(pImgs[i], img2_obj, cv2.NORM_L2) / len(img2_obj)
    mean_error += ierror
mean_error = mean_error / len(pObjs)
print('------mean error is: ', mean_error)