Hegong sky team vision training day4 - traditional vision, contour recognition

Catalog

Learning goals

Learning content

Learning time

Learning output

One 、findContours Function learning notes

Two 、 Look for the light bar on the armor plate

1. call findContours Find function outline , The functions used inside are as follows findContours、drawContours、minAreaRect、contourArea It's all in csdn Checked on , I won't elaborate here .

2. Traverse all contours

3. Screen out the light bar on the armor plate

3、 ... and 、 Find the center

The learning

---

Learning goals

• Contour recognition

---

Learning content

1. findContours function
2. Look for the light bar
3. Find the center

---

Learning time

• 2022 year 7 month 21 Japan

---

Learning output

One 、findContours Function learning notes

 Contour extraction uses findContours function , Its parameters are as follows ：

void cv::findContours	(	InputArray 
image,
		OutputArrayOfArrays 
contours,
		OutputArray 
hierarchy,
		int 	mode,
		int 	method,
		Point 
offset = Point() 

	)		

void cv::findContours	(	InputArray 
image,
		OutputArrayOfArrays 
contours,
		int 	mode,
		int 	method,
		Point 
offset = Point() 

	)	
	

First is the understanding of function declaration and the analysis of team training materials ：

image： The input image .8-bit Single channel binary image , Non zero pixels are treated as 1.

contours： Detected contours . It's a vector , Every element of a vector is an outline . therefore , Every element of this vector is still a vector . namely vector<vector<Point> > contours

hierarchy： The inheritance relationship of each outline .hierarchy It's also a vector , Length and contours equal , Each element and contours The elements of are corresponding to .hierarchy Each element of is a vector of four integers . namely ：            vector<Vec4i> hierarchy; //Vec4i is a vector contains four number of int

int Type mode, Define the retrieval mode of the contour ：

Take one ：CV_RETR_EXTERNAL Only the outermost contour is detected , The inner contour contained within the outer contour is ignored ;

Take two ：CV_RETR_LIST   Detect all contours , Including the inner circumference 、 The outer contour , However, the detected contour does not establish a hierarchical relationship ;

Take three ：CV_RETR_CCOMP   Detect all contours , But all contours establish only two hierarchical relationships , The periphery is the top floor , If the inner contour in the periphery also contains other contour information , Then all contours in the inner wall belong to the top layer ;

Take four ：CV_RETR_TREE   Detect all contours , All contours establish a hierarchical tree structure . The outer contour contains the inner contour , The inner contour can also continue to contain embedded contours .

Fifth parameter ：int Type method, Approximate method of defining contour ：

Take one ：CV_CHAIN_APPROX_NONE Save all continuous contour points on the object boundary to contours Within vector

Take two ：CV_CHAIN_APPROX_SIMPLE  Save only the inflection point information of the contour , Save all points at the inflection point of the contour into contours Within vector , The information points on the straight line between the inflection point and the inflection point will not be retained

among
（1）hierarchy Represents the hierarchical relationship of the contour , For the first i ii Strip outline ,hierarchy[i][0] , hierarchy[i][1] , hierarchy[i][2] , hierarchy[i][3] Respectively represent the next contour 、 The previous outline 、（ The first one at the same level ） Sub contour 、 Index of the parent contour （ If there is no corresponding index , It's negative ）.

（2）mode parameter “ Contour retrieval mode （Contour retrieval mode）”, Contains RETR_LIST,RETR_EXTERNAL,RETR_CCOMP,RETR_TREE Four patterns .

（3）method Parameters represent contour representation , It is generally used CHAIN_APPROX_SIMPLE.

After learning, I have an understanding of functions , But it has not formed an intuitive impression , Need to learn more .

Two 、 Look for the light bar on the armor plate

1. call findContours Find function outline , The functions used inside are as follows findContours、drawContours、minAreaRect、contourArea It's all in csdn Checked on , I won't elaborate here .

2. Traverse all contours

After traversing all the contours, draw all the recognized contours in the original drawing , To avoid other contour interference , Use minAreaRect Functions and contourArea Function to get a series of information about the smallest circumscribed rectangle of all contours , And through the geometric properties of other minimum circumscribed rectangles , Screen out the light bar on the armor plate .

3. Screen out the light bar on the armor plate

Use minAreaRect Function to get the center coordinate of the smallest circumscribed rectangle 、 Long 、 wide 、 Rotation Angle , Get the area of the smallest circumscribed rectangle, and you can filter out the rectangle of the light bar .

The code is as follows ：

import cv2
import numpy as np

img = cv2.imread('/ Picture path /')  
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)		

contours,Hierarchy = cv2.findContours(binary,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_NONE)
print (type(contours))  
print (type(contours[0])) 

>>> <class 'list'>				
	<class 'numpy.ndarray'>		
for cnt in range(contours[cnt]):  
    cv2.drawContours(frame, contours[cnt],-1,(255,0,0),3)           # Draw the outline 
rect = cv2.minAreaRect(cnt)

print(" Central coordinates ：", rect[0])
print(" Width ：", rect[1][0])
print(" length ：", rect[1][1])
print(" Rotation Angle ：", rect[2])
    
area = cv2.contourArea(contours[cnt])

print(" area ：", rect[0])

But at present, I only deal with binary images , But there are so many contours of such binary pictures , It's a little difficult for later matching , Because our task is to give a picture , And there are not many influencing factors , The light bar can be identified .

3、 ... and 、 Find the center

It's really difficult for me to write this part by myself , I refer to some codes of Southeast University and my own understanding, but the code still has bug, The idea needs to be improved .

The general idea is to search for the contour after image binarization （ By calculating the area ）, Then screen the outline from the length width ratio of the light bar and mark the center , Finally, select it and draw a circle in the center .

PS: The code is partly extracted from the open source project file of Southeast University , So it's just a reflection of my own thoughts , Actually, it should not work , It includes part of identifying contour and finding center point , Compared with my own recognition contour , Dongda's code is many times stronger than mine , I've also watched it for a long time to learn this whole set of ideas , Now we can only search after binarization , And draw a circle , Understanding may be biased , Now I'm immature , Still need follow-up learning to improve .

#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <opencv2/opencv.hpp>
#include <iostream>
#include <string>
#include<vector>
#include <time.h>
using namespace cv;
using namespace std;
 
const int ANGLE_TO_UP = 1;
const int WIDTH_GREATER_THAN_HEIGHT = 0;
 

    vector<RotatedRect> vc;
    vector<RotatedRect> vRec;
    vector<vector<Point>> Light_Contour;
    findContours(element.clone(),Light_Contour, RETR_EXTERNAL,CHAIN_APPROX_SIMPLE);

    for (int i = 0;i < Light_Contour.size();i++)
    {
        float Light_Contour_Area = contourArea(Light_Contour[i]);
        if (Light_Contour_Area < 15 || Light_Contour[i].size() <= 10)
            continue;
        RotatedRect Light_Rec = fitEllipse(Light_Contour[i]);
        Light_Rec = adjustRec(Light_Rec, ANGLE_TO_UP);
 
        if (Light_Rec.angle > 10 )
        if (Light_Rec.size.width / Light_Rec.size.height > 1.5
                || Light_Contour_Area / Light_Rec.size.area() < 0.5)
            continue;
        Light_Rec. size.height *= 1.1;
        Light_Rec.size.width *= 1.1;
        vc.push_back(Light_Rec);
    }
    for (size_t i = 0; i < vc.size(); i++)
    {
        for (size_t j = i + 1; (j < vc.size()); j++)
        {
            float Contour_angle = abs(vc[i].angle - vc[j].angle); // Angle difference 
            if (Contour_angle >= 7)
            float Contour_Len1 = abs(vc[i].size.height - vc[j].size.height) / max(vc[i].size.height, vc[j].size.height);
            float Contour_Len2 = abs(vc[i].size.width - vc[j].size.width) / max(vc[i].size.width, vc[j].size.width);
            if (Contour_Len1 > 0.25 || Contour_Len2 > 0.25)
                continue;
 
            RotatedRect ARMOR;
            ARMOR.center.x = (vc[i].center.x + vc[j].center.x) / 2.; 
            ARMOR.center.y = (vc[i].center.y + vc[j].center.y) / 2.; 
            ARMOR.angle = (vc[i].angle + vc[j].angle) / 2.; 
            float nh, nw, yDiff, xDiff;
            nw = sqrt((vc[i].center.x - vc[j].center.x) * (vc[i].center.x - vc[j].center.x) + (vc[i].center.y - vc[j].center.y) * (vc[i].center.y - vc[j].center.y));
            float ratio = nw / nh; 
            xDiff = abs(vc[i].center.x - vc[j].center.x) / nh; 
            yDiff = abs(vc[i].center.y - vc[j].center.y) / nh; 
            if (ratio < 1.0 || ratio > 5.0 || xDiff < 0.5 || yDiff > 2.0)
                continue;
            ARMOR.size.height = nh;
            ARMOR.size.width = nw;
            vRec.push_back(ARMOR);
            Point2f point1;
            Point2f point2;
            point1.x=vc[i].center.x;point1.y=vc[i].center.y+20;
            point2.x=vc[j].center.x;point2.y=vc[j].center.y-20;
            int xmidnum = (point1.x+point2.x)/2;
            int ymidnum = (point1.y+point2.y)/2;
            ARMOR.center.x = filter(ARMOR.center.x,xmidnum, DELAT_MAX);
            ARMOR.center.y = filter(ARMOR.center.y,ymidnum, DELAT_MAX);
            Scalar color(rand() & 255, rand() & 255, rand() & 255);
            rectangle(imgOriginal, point1,point2, color, 2);
            circle(imgOriginal,ARMOR.center,10,color);
 
        }
    }

---

The learning

I feel that I still have a lot of knowledge to learn , There are many problems in the compilation process bug, such as error C4430: Missing type specifier - Is assumed to be int. Be careful : C++ Default... Is not supported int wait , I don't have a thorough knowledge of many functions , In the third step, we refer to the code of Dongda and find that the whole set is not of the same order of magnitude , Now I'm just a cute new , The road to learning is heavy and long .