采用opencv 对图像中的小球通过canny边缘检测的方式进行提取坐标
本文介绍了如何使用OpenCV对图像中的小球进行Canny边缘检测，并通过Zernike矩进行亚像素边缘检测，最终拟合椭圆以获取小球的精确坐标。首先，图像被转换为灰度图并进行高斯平滑处理，接着通过Canny算法检测边缘。然后，利用Zernike矩对边缘点进行亚像素级别的精确定位，最后通过椭圆拟合得到小球的中心坐标。该方法能够有效提高小球检测的精度，适用于需要高精度定位的图像处理任务。
cpp
#include <opencv2/opencv.hpp>
#include <iostream>
#include <vector>
#include <map>
#include <cmath>
#pragma comment(linker, "/subsystem:\"console\" /entry:\"mainCRTStartup\"" )
//#pragma comment(linker, "/subsystem:\"windows\" /entry:\"mainCRTStartup\"")
// 全局变量存储显示状态
double scale = 1.0; // 当前缩放比例
int offset_x = 0; // X轴偏移量
int offset_y = 0; // Y轴偏移量
cv::Point mouse_pt(-1, -1); // 鼠标位置
bool dragging = false; // 拖拽状态
struct ZernikeTemplates {
cv::Mat real;
cv::Mat imag;
cv::Mat mod;
};
ZernikeTemplates generateZernikeTemplates(int N) {
ZernikeTemplates zt;
int size = 2 * N + 1;
zt.real = cv::Mat::zeros(size, size, CV_64F);
zt.imag = cv::Mat::zeros(size, size, CV_64F);
zt.mod = cv::Mat::zeros(size, size, CV_64F);
for (int y = -N; y <= N; ++y) {
for (int x = -N; x <= N; ++x) {
double rho = sqrt(x * x + y * y) / N;
double theta = atan2(y, x);
if (rho <= 1.0) {
zt.real.at<double>(y + N, x + N) = rho * cos(theta);
zt.imag.at<double>(y + N, x + N) = rho * sin(theta);
//zt.mod.at<double>(y + N, x + N) = 1.0;
}
else
zt.mod.at<double>(y + N, x + N) = 1.0;
}
}
// 归一化
double sum_real = sum(abs(zt.real))[0];
double sum_imag = sum(abs(zt.imag))[0];
double sum_mod = sum(zt.mod)[0];
zt.real /= sum_real;
zt.imag /= sum_imag;
zt.mod /= sum_mod;
return zt;
}
std::vector<cv::Point2d> zernikeEdgeDetection(const cv::Mat& grayImg, const cv::Mat& edges,int zernikeN,double minEdgeStrength)
{
// 生成Zernike模板
ZernikeTemplates zt = generateZernikeTemplates(zernikeN);
// 收集边缘点
std::vector<cv::Point> edgePixels;
findNonZero(edges, edgePixels);
// 亚像素边缘检测
std::vector<cv::Point2d> subPixelEdges;
for (const cv::Point& p : edgePixels) {
int x = p.x;
int y = p.y;
// 检查边界
if (x < zernikeN || x >= grayImg.cols - zernikeN ||
y < zernikeN || y >= grayImg.rows - zernikeN)
continue;
// 提取ROI
cv::Rect roi(x - zernikeN, y - zernikeN, 2 * zernikeN + 1, 2 * zernikeN + 1);
cv::Mat region = grayImg(roi).clone();
region.convertTo(region, CV_64F, 1.0 / 255.0);
// 计算Zernike矩
double M_real = sum(region.mul(zt.real))[0];
double M_imag = sum(region.mul(zt.imag))[0];
double M_mod = sum(region.mul(zt.mod))[0];
// 计算边缘参数
double theta = atan2(M_imag, M_real);
double edgeStrength = sqrt(M_real * M_real + M_imag * M_imag) / M_mod;
// 计算偏移量
double l = (3 * M_mod - 4 * (M_real * cos(theta) + M_imag * sin(theta)))
/ (2 * edgeStrength + 1e-6);
// 有效性检查
if (abs(l) > zernikeN * 1.5 || edgeStrength < minEdgeStrength)
continue;
// 计算亚像素坐标
cv::Point2d subPixel(p.x + l * cos(theta)+1, p.y + l * sin(theta)+1); // MARLAB坐标偏移补偿
subPixelEdges.push_back(subPixel);
}
return subPixelEdges;
}
cv::RotatedRect fitEllipseRobust(const std::vector<cv::Point2d>& points) {
// 将点转换为Point2f格式
std::vector<cv::Point2f> pointsF;
int i = 0;
for (const auto& p : points)
{
i++;
pointsF.emplace_back(p.x, p.y);
std::cout << i << " point:" << p.x << " " << p.y << std::endl;
}
// 使用OpenCV的椭圆拟合（带鲁棒性）
cv::RotatedRect ellipse = fitEllipse(pointsF);
return ellipse;
}
int main()
{
// 1.读取bmp图片及显示结果
cv::Mat image = cv::imread("G:/tools/matlab2020b/Zernike/left_4.bmp",cv::IMREAD_COLOR);
if (image.empty()) {
std::cout << "Could not open or find the image" << std::endl;
return -1;
}
//cv::imshow("原始图像", image);
// 2.进行Canny阈值边缘检测及显示结果
// 参数设置
double canny_thresh[2] = { 0.1, 0.4 }; // Canny阈值
int zernike_N = 7;
double min_edge_strength = 0.1;
cv::Mat gray_img;
// 转换为灰度图
if (image.channels() == 3) {
cv::cvtColor(image, gray_img, cv::COLOR_BGR2GRAY);
}
else {
gray_img = image.clone();
}
// 归一化处理，类似于MATLAB的im2double
gray_img.convertTo(gray_img, CV_64F, 1.0 / 255.0);
// 转换回CV_8U类型
cv::Mat gray_img_8u;
gray_img.convertTo(gray_img_8u, CV_8U, 255.0);
// 手动对图像进行高斯平滑处理
cv::Mat blurred;
cv::GaussianBlur(gray_img_8u, blurred, cv::Size(5,5),0.3);// MATLAB默认sigma=0.6
//cv::GaussianBlur(gray_img_8u, blurred, cv::Size(3, 3), 0);
// Otsu 阈值分割
cv::Mat binary;
cv::threshold(blurred, binary, 0, 255, cv::THRESH_BINARY + cv::THRESH_OTSU);
// 步骤1: Canny边缘检测
cv::Mat BW;
double low_thresh = canny_thresh[0] * 255;
double high_thresh = canny_thresh[1] * 255;
cv::Canny(binary * 255, BW, low_thresh, high_thresh, 7);
cv::imshow("Canny图像", BW);
cv::imwrite("G:/tools/matlab2020b/Zernike/Canny.bmp", BW);
// 3.Zernike矩亚像素边缘检测及显示结果
std::vector<cv::Point2d> subPixelEdges = zernikeEdgeDetection(binary, BW, zernike_N, min_edge_strength);
// 椭圆拟合
if (subPixelEdges.size() < 5) {
std::cerr << "Not enough points for ellipse fitting" << std::endl;
return -1;
}
cv::RotatedRect ellipse = fitEllipseRobust(subPixelEdges);
// 显示结果
cv::Mat display = image.clone();
// 绘制边缘点
for (const auto& p : subPixelEdges)
{
circle(display, p, 0.1, cv::Scalar(0, 255, 255), cv::FILLED);
}
// 绘制椭圆
//cv::ellipse(display, ellipse, cv::Scalar(0, 0, 255), 1);
// 绘制中心
circle(display, ellipse.center, 0.1, cv::Scalar(0, 0, 255), cv::FILLED);
std::cout << "-------center-------" << ellipse.center.x << " " << ellipse.center.y << std::endl;
// 创建窗口并设置回调
cv::namedWindow("Zoomable Window", cv::WINDOW_AUTOSIZE);
cv::imshow("Zoomable Window", display);
// 4.椭圆拟合及显示结果
cv::imwrite("G:/tools/matlab2020b/Zernike/edges.bmp", display);
//cv::Mat combined;
//cv::hconcat(image, CannyImage, combined);
//cv::imshow("图像", combined);
cv::waitKey(0);
}matLab中的提取像素坐标序列从1开始 而 opencv是从0开始，所以坐标相差1是正常的 本文中为了像matLab对齐，采用了补全+1的方式。
如下为cmakeLists
cpp
cmake_minimum_required(VERSION 3.0)
project(CanDy_Demo)
set(CMAKE_AUTOUIC ON)
set(CMAKE_AUTOMOC ON)
set(CMAKE_AUTORCC ON)
include_directories(
${PROJECT_SOURCE_DIR}
${PROJECT_SOURCE_DIR}/opencv4/include
${PROJECT_BINARY_DIR})
set(srcs
main.cpp
)
set(hdrs
)
set(uis
)
add_executable(${PROJECT_NAME} ${srcs} ${hdrs} ${wrapUis})
add_compile_definitions(ARMDCOMBINEDAPI_EXPORTS)
add_compile_definitions(NOMINMAX)
#opencv libss
find_library(OPENCV_WORLD_LIBRARY opencv_world452 ${CMAKE_CURRENT_SOURCE_DIR}/Opencv4/lib)
target_link_libraries(${PROJECT_NAME}
#opencv lib
${OPENCV_WORLD_LIBRARY}
)
if(EXISTS ${CMAKE_SOURCE_DIR}/${DIRS_IN_SRC})
execute_process(COMMAND ${CMAKE_COMMAND} -E copy_directory
${CMAKE_SOURCE_DIR}/${DIRS_IN_SRC}
${CMAKE_CURRENT_BINARY_DIR}/${DIRS_IN_SRC})
endif()
#SET_TARGET_PROPERTIES(${PROJECT_NAME} PROPERTIES LINK_FLAGS "/MANIFESTUAC:\"level='requireAdministrator' uiAccess='false'\" /SUBSYSTEM:WINDOWS")