OpenCV 相机标定流程指南

• OpenCV 相机标定流程指南
  • 前置准备
  • 标定流程
  • 结果输出与验证
  • 建议
  • 源代码

OpenCV 相机标定流程指南

https://docs.opencv.org/4.x/dc/dbb/tutorial_py_calibration.html
https://learnopencv.com/camera-calibration-using-opencv/

前置准备

1. 制作标定板：生成高精度棋盘格或圆点标定板。
2. 采集标定板图像：在不同角度、距离和光照条件下采集多张标定板图像。

OpenCV 官方标定板生成脚本使用教程
!OpenCV 官方标定板脚本下载

访问我的源代码仓库下载已经生成的矢量棋盘网格，使用打印机打印出来即可进行图像标定采集工作。

标定流程

使用 CameraCalib 类进行相机标定：

1. 添加图像样本：将采集的标定板图像导入标定系统。
2. 并发检测角点：利用多线程技术并行检测图像中的角点或特征点。
3. 相机标定：基于检测到的角点，计算相机内参（焦距、主点坐标）和外参（旋转矩阵、平移向量），并优化畸变系数。

结果输出与验证

1. 打印标定结果：输出相机内参、外参及畸变系数。
2. 测试图像标定：使用标定结果对测试图像进行畸变校正，验证标定精度。

建议

可信误差：重投影误差应小于 0.5 像素，最大不超过 1.0 像素。
采集夹角要求：摄像头与标定板平面的夹角应控制在 30°~60° 之间，避免极端角度。

[1] https://www.microsoft.com/en-us/research/publication/a-flexible-new-technique-for-camera-calibration/

源代码

#include <opencv2/opencv.hpp>
#include <algorithm>
#include <memory>
#include <vector>
#include <string>
#include <print>
#include <iostream>class CameraCalib
{
public:// 校准模式enum class Pattern : uint32_t {CALIB_SYMMETRIC_CHESSBOARD_GRID,  // 规则排列的棋盘网格 // chessboardCALIB_MARKER_CHESSBOARD_GRID,     // 额外标记的棋盘网格 // marker chessboardCALIB_SYMMETRIC_CIRCLES_GRID,     // 规则排列的圆形网格 // circlesCALIB_ASYMMETRIC_CIRCLES_GRID,    // 交错排列的圆形网格 // acirclesCALIB_PATTERN_COUNT,              // 标定模式的总数量 用于 for 循环遍历 std::to_underlying(Pattern::CALIB_PATTERN_COUNT);};struct CameraCalibrationResult {cv::Mat cameraMatrix;                     // 相机矩阵（内参数）cv::Mat distortionCoefficients;           // 畸变系数double reprojectionError;                 // 重投影误差（标定精度指标）std::vector<cv::Mat> rotationVectors;     // 旋转向量（外参数）std::vector<cv::Mat> translationVectors;  // 平移向量（外参数）};explicit CameraCalib(int columns, int rows, double square_size /*mm*/, Pattern pattern): patternSize_(columns, rows), squareSize_(square_size), pattern_(pattern) {// 构造一个与标定板对应的真实的世界角点数据for(int y = 0; y < patternSize_.height; ++y) {for(int x = 0; x < patternSize_.width; ++x) {realCorners_.emplace_back(x * square_size, y * square_size, 0.0f);}}}void addImageSample(const cv::Mat &image) { samples_.emplace_back(image); }void addImageSample(const std::string &filename) {cv::Mat mat = cv::imread(filename, cv::IMREAD_COLOR);if(mat.empty()) {std::println(stderr, "can not load filename: {}", filename);return;}addImageSample(mat);}bool detectCorners(const cv::Mat &image, std::vector<cv::Point2f> &corners) {bool found;switch(pattern_) {using enum Pattern;case CALIB_SYMMETRIC_CHESSBOARD_GRID: detectSymmetricChessboardGrid(image, corners, found); break;case CALIB_MARKER_CHESSBOARD_GRID: detectMarkerChessboardGrid(image, corners, found); break;case CALIB_SYMMETRIC_CIRCLES_GRID: detectSymmetricCirclesGrid(image, corners, found); break;case CALIB_ASYMMETRIC_CIRCLES_GRID: detectAsymmetricCirclesGrid(image, corners, found); break;default: break;}return found;}std::vector<std::vector<cv::Point2f>> detect() {std::vector<std::vector<cv::Point2f>> detectedCornerPoints;std::mutex mtx;  // 使用 mutex 来保护共享资源std::atomic<int> count;std::for_each(samples_.cbegin(), samples_.cend(), [&](const cv::Mat &image) {std::vector<cv::Point2f> corners;bool found = detectCorners(image, corners);if(found) {count++;std::lock_guard<std::mutex> lock(mtx);  // 使用 lock_guard 来保护共享资源detectedCornerPoints.push_back(corners);}});std::println("Detection successful: {} corners, total points: {}", int(count), detectedCornerPoints.size());return detectedCornerPoints;}std::unique_ptr<CameraCalibrationResult> calib(std::vector<std::vector<cv::Point2f>> detectedCornerPoints, int width, int height) {// 准备真实角点的位置std::vector<std::vector<cv::Point3f>> realCornerPoints;for(size_t i = 0; i < detectedCornerPoints.size(); ++i) {realCornerPoints.emplace_back(realCorners_);}cv::Size imageSize(width, height);// 初始化相机矩阵和畸变系数cv::Mat cameraMatrix = cv::Mat::eye(3, 3, CV_64F);cv::Mat distCoeffs   = cv::Mat::zeros(5, 1, CV_64F);std::vector<cv::Mat> rvecs, tvecs;// 进行相机标定double reproError = cv::calibrateCamera(realCornerPoints, detectedCornerPoints, imageSize, cameraMatrix, distCoeffs, rvecs, tvecs, cv::CALIB_FIX_K1 + cv::CALIB_FIX_K2 + cv::CALIB_FIX_K3 + cv::CALIB_FIX_K4 + cv::CALIB_FIX_K5);// 将标定结果存储到结构体中auto result                    = std::make_unique<CameraCalibrationResult>();result->cameraMatrix           = cameraMatrix;result->distortionCoefficients = distCoeffs;result->reprojectionError      = reproError;result->rotationVectors        = rvecs;result->translationVectors     = tvecs;return result;}// 打印标定结果void print(const std::unique_ptr<CameraCalibrationResult> &result) {std::cout << "重投影误差: " << result->reprojectionError << std::endl;std::cout << "相机矩阵:\n" << result->cameraMatrix << std::endl;std::cout << "畸变系数:\n" << result->distortionCoefficients << std::endl;}// 进行畸变校正测试void test(const std::string &filename, const std::unique_ptr<CameraCalibrationResult> &param) {// 读取一张测试图像cv::Mat image = cv::imread(filename);if(image.empty()) {std::println("can not load filename");return;}cv::Mat undistortedImage;cv::undistort(image, undistortedImage, param->cameraMatrix, param->distortionCoefficients);// 显示原图和校准后的图cv::namedWindow("Original Image", cv::WINDOW_NORMAL);cv::namedWindow("Undistorted Image", cv::WINDOW_NORMAL);cv::imshow("Original Image", image);cv::imshow("Undistorted Image", undistortedImage);// 等待用户输入任意键cv::waitKey(0);}private:void dbgView(const cv::Mat &image, const std::vector<cv::Point2f> &corners, bool &found) {if(!found) {std::println("Cannot find corners in the image");}// Debug and view detected corner points in imagesif constexpr(false) {cv::drawChessboardCorners(image, patternSize_, corners, found);cv::namedWindow("detectCorners", cv::WINDOW_NORMAL);cv::imshow("detectCorners", image);cv::waitKey(0);cv::destroyAllWindows();}}void detectSymmetricChessboardGrid(const cv::Mat &image, std::vector<cv::Point2f> &image_corners, bool &found) {if(found = cv::findChessboardCorners(image, patternSize_, image_corners); found) {cv::Mat gray;cv::cvtColor(image, gray, cv::COLOR_BGR2GRAY);cv::cornerSubPix(gray, image_corners, cv::Size(11, 11), cv::Size(-1, -1), cv::TermCriteria(cv::TermCriteria::EPS + cv::TermCriteria::COUNT, 30, 0.01));dbgView(image, image_corners, found);}}void detectMarkerChessboardGrid(const cv::Mat &image, std::vector<cv::Point2f> &image_corners, bool &found) {if(found = cv::findChessboardCornersSB(image, patternSize_, image_corners); found) {dbgView(image, image_corners, found);}}void detectSymmetricCirclesGrid(const cv::Mat &image, std::vector<cv::Point2f> &image_corners, bool &found) {if(found = cv::findCirclesGrid(image, patternSize_, image_corners, cv::CALIB_CB_SYMMETRIC_GRID); found) {cv::Mat gray;cv::cvtColor(image, gray, cv::COLOR_BGR2GRAY);cv::cornerSubPix(gray, image_corners, cv::Size(11, 11), cv::Size(-1, -1), cv::TermCriteria(cv::TermCriteria::EPS + cv::TermCriteria::COUNT, 30, 0.01));dbgView(image, image_corners, found);}}void detectAsymmetricCirclesGrid(const cv::Mat &image, std::vector<cv::Point2f> &image_corners, bool &found) {cv::SimpleBlobDetector::Params params;params.minThreshold = 8;params.maxThreshold = 255;params.filterByArea = true;params.minArea      = 50;    // 适当降低，以便检测小圆点params.maxArea      = 5000;  // 适当降低，以避免误检大区域params.minDistBetweenBlobs = 10;  // 调小以适应紧密排列的圆点params.filterByCircularity = false;  // 允许更圆的形状params.minCircularity      = 0.7;    // 只有接近圆的目标才被识别params.filterByConvexity = true;params.minConvexity      = 0.8;  // 只允许较凸的形状params.filterByInertia = true;params.minInertiaRatio = 0.1;  // 适应不同形状params.filterByColor = false;  // 关闭颜色过滤，避免黑白检测问题auto blobDetector = cv::SimpleBlobDetector::create(params);if(found = cv::findCirclesGrid(image, patternSize_, image_corners, cv::CALIB_CB_ASYMMETRIC_GRID | cv::CALIB_CB_CLUSTERING, blobDetector); found) {cv::Mat gray;cv::cvtColor(image, gray, cv::COLOR_BGR2GRAY);cv::cornerSubPix(gray, image_corners, cv::Size(11, 11), cv::Size(-1, -1), cv::TermCriteria(cv::TermCriteria::EPS + cv::TermCriteria::COUNT, 30, 0.01));dbgView(image, image_corners, found);}}private:cv::Size patternSize_;double squareSize_;Pattern pattern_;std::vector<cv::Point3f> realCorners_;std::vector<cv::Mat> samples_;
};// 测试函数
static void test_CameraCalib() {// 创建一个 CameraCalib 对象，指定标定板大小、每个方格的边长和校准模式CameraCalib calib(14, 9, 12.1, CameraCalib::Pattern::CALIB_MARKER_CHESSBOARD_GRID);// 加载图像样本std::vector<cv::String> result;cv::glob("calibration_images/*.png", result, false);for (auto &&filename : result) {calib.addImageSample(filename);}// 检测角点auto detectedCornerPoints = calib.detect();// 进行相机标定std::string filename = "calibration_images/checkerboard_radon.png";cv::Mat image = cv::imread(filename);if (image.empty()) {std::println("can not load image");return;}auto param = calib.calib(detectedCornerPoints, image.cols, image.cols);// 打印标定结果calib.print(param);// 测试函数calib.test(filename, param);
}

运行测试函数，输出结果如下所示：

Detection successful: 2 corners, total points: 2
重投影误差: 0.0373256
相机矩阵:
[483030.3184975122, 0, 1182.462802265994;0, 483084.13533141, 1180.358683128085;0, 0, 1]
畸变系数:
[0;0;-0.002454905573938355;9.349667940808669e-05;0]

 // 保存标定结果
cv::FileStorage fs("calibration_result.yml", cv::FileStorage::WRITE);
fs << "camera_matrix" << result.cameraMatrix;
fs << "distortion_coefficients" << result.distCoeffs;
fs << "image_size" << result.imageSize;
fs.release();