OpenCV中的分段错误

Question

我最近开始使用C++来做一些图像处理，并面临分割错误的问题。希望有人能帮我弄清楚发生了什么事？谢谢问题是，下面所附的代码运行良好，但如果我的"；向量roi_ corners(4("变为"；向量roi_ corners"则使用"；push_back(("以及"；clear(("；对向量进行更新，会产生分割错误。有人能帮我澄清这个问题的原因吗？非常感谢。

OpenCV版本：4.4.0

MacOS版本：10.14.5

可工作代码

#include <opencv2/opencv.hpp>
#include <iostream>
#include <math.h>
#include <vector>  
#define PI 3.14159265
#define WINDOW "Image Correction"
using namespace std;
using namespace cv;
vector< Point2f> roi_corners(4); 
vector< Point2f> dst_corners(4); 
Mat img1;
Mat cache;
int roi_id;
void On_mouse(int event, int x, int y, int flags, void*);

int main(int argc, char const *argv[])
{

//import image
roi_id = 0;
img1 = imread("Board.jpg");
if ( img1.empty() )
{
cerr << "Please Import an Image!!" << endl;
}
int factor = 60;//for pixel adjustment
cache = img1.clone();//copy for retake points
imshow(WINDOW, img1);


/*Run the point taking procedure*/
while(true){
setMouseCallback(WINDOW, On_mouse, 0);
char c = (char)waitKey( 10 );
if(c=='n') break;//press 'n' when determine the four point you want 
if(c=='e') {roi_id=0; img1 = cache.clone();} //press 'e' to retake the foru point
}

/*For adjustment point estimation*/
dst_corners[0].x = roi_corners[0].x;
dst_corners[0].y = roi_corners[0].y;
dst_corners[1].x = roi_corners[0].x+factor*1;
dst_corners[1].y = roi_corners[0].y;
dst_corners[2].x = roi_corners[0].x+factor*1;
dst_corners[2].y = roi_corners[0].y+factor*1;
dst_corners[3].x = roi_corners[0].x;
dst_corners[3].y = roi_corners[0].y+factor*1;
Mat M = getPerspectiveTransform(roi_corners, dst_corners);
Mat warped_image;

/*Print the corrected picture*/
Size sz = cache.size();
warpPerspective(cache, warped_image, M, Size(sz.width, sz.height)); // do perspective transformation
imshow("Corrected Image", warped_image);
waitKey(0);
cout<<"complete"<<endl;
return 0;
}

void On_mouse(int event, int x, int y, int flags, void*)
{   imshow(WINDOW, img1);
if(roi_id<4){
if (event == EVENT_LBUTTONDOWN){
roi_corners[roi_id].x=x; 
roi_corners[roi_id].y=y;
cout<<"The Point You Take is: "<<x<<' '<<y<<endl;
roi_id++;
circle(img1, Point(x,y), 2, Scalar(0, 0, 255), LINE_8 ,0);
imshow(WINDOW, img1);
}
}
}

如果我这样修改代码，它将显示seg故障

#include <opencv2/opencv.hpp>
#include <iostream>
#include <math.h>
#include <vector>  
#define PI 3.14159265
#define WINDOW "Image Correction"
using namespace std;
using namespace cv;
vector< Point2f> roi_corners; 
vector< Point2f> dst_corners; 
Mat img1;
Mat cache;

void On_mouse(int event, int x, int y, int flags, void*);

int main(int argc, char const *argv[])
{

//import image
roi_id = 0;
img1 = imread("Board.jpg");
if ( img1.empty() )
{
cerr << "Please Import an Image!!" << endl;
}
int factor = 60;//for pixel adjustment
cache = img1.clone();//copy for retake points
imshow(WINDOW, img1);


/*Run the point taking procedure*/
while(true){
setMouseCallback(WINDOW, On_mouse, 0);
char c = (char)waitKey( 10 );
if(c=='n') break;//press 'n' when determine the four point you want 
if(c=='e') {roi_corner.clear(); img1 = cache.clone();} //press 'e' to retake the foru point
}

/*For adjustment point estimation*/
dst_corners[0].x = roi_corners[0].x;
dst_corners[0].y = roi_corners[0].y;
dst_corners[1].x = roi_corners[0].x+factor*1;
dst_corners[1].y = roi_corners[0].y;
dst_corners[2].x = roi_corners[0].x+factor*1;
dst_corners[2].y = roi_corners[0].y+factor*1;
dst_corners[3].x = roi_corners[0].x;
dst_corners[3].y = roi_corners[0].y+factor*1;
Mat M = getPerspectiveTransform(roi_corners, dst_corners);
Mat warped_image;

/*Print the corrected picture*/
Size sz = cache.size();
warpPerspective(cache, warped_image, M, Size(sz.width, sz.height)); // do perspective transformation
imshow("Corrected Image", warped_image);
waitKey(0);
cout<<"complete"<<endl;
return 0;
}

void On_mouse(int event, int x, int y, int flags, void*)
{   imshow(WINDOW, img1);
if(roi_corners.size()<4){
if (event == EVENT_LBUTTONDOWN){
roi_corners.push_back(Point2f(x,y)); 
cout<<"The Point You Take is: "<<x<<' '<<y<<endl;
circle(img1, Point(x,y), 2, Scalar(0, 0, 255), LINE_8 ,0);
imshow(WINDOW, img1);
}
}
}

Answer 1

代码中的一些内容。您声明了一个由4个元素组成的向量，但没有对其进行初始化。根据您的平台和数据类型，您可能会看到不需要的行为。尝试在同一行声明和初始化向量：

std::vector<cv::Point2f> roi_corners( 4, cv::Point2f(0.0, 0.0) );

当然，如果您用初始容量(size(和初始值声明向量，则可以使用std::vector<>::operator[]:对每个元素进行索引

roi_corners[0] = cv::Point2f( 1.0, 2.0 );
roi_corners[1] = cv::Point2f( 3.0, 4.0 );
roi_corners[2] = cv::Point2f( 5.0, 6.0 );
roi_corners[3] = cv::Point2f( 7.0, 8.0 );

通过声明具有初始大小的向量，您已经分配了内存，这些内存将用于向量的单个元素store和load。现在，假设您没有用初始大小声明向量，并使用push_back添加元素：

//vector declaration with no initial size:
std::vector<cv::Point2f> roi_corners;
//store a new element into the vector:
roi_corners.push_back( cv::Point2f(1.0, 2.0) );

很酷，你的矢量已经存储了一个新元素，并且只显示了一个项目的容量。然而，你仍然这样做：

roi_corners[0] = cv::Point2f( 1.0, 2.0 ); // data overwrite in position 0
roi_corners[1] = cv::Point2f( 3.0, 4.0 ); // you haven't allocated memory for this yet!
Result: seg fault

推论：如果您有一个预定义的大小为N的向量，您可以通过std::vector<>::operator[]将元素从0索引到N，因为内存已分配用于容纳所有N元素。如果您尝试寻址此范围之外的元素，则会收到一个seg fault。