基于单目的Opencv的线径、像素尺寸、线宽、目标尺寸计算

引言

想法起源于控制3d打印机初始线径，能够监控平台首层高度是否合理。经过实际测量，不同的高度，精度需求为0.01mm，即程序应该能够区分之分辨率。

```python
%load_ext autoreload
%autoreload 2
import cv2
import os 
import numpy as np
from utils import objectFound
```

```python
imgs = [os.path.join("../images/",x) for x in os.listdir("../images/") if x.find("re")>-1]
imgs
```

    ['../images/1re.jpg', '../images/2re.jpg', '../images/3re.jpg']

## 测量精度
* 单个2cm方块精度: 0.024

```python
img1 = cv2.imread(imgs[1])
img1.shape
```

    (1840, 3264, 3)

```python
gray = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
blur = cv2.GaussianBlur(gray, (9,9), 0)
# _, thresh = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
_, thresh = cv2.threshold(blur, 150, 255, 0)
cv2.namedWindow("thresh", cv2.WINDOW_NORMAL)
cv2.imshow("thresh", thresh)
cv2.waitKey(0)
cv2.destroyAllWindows()
```

```python
ob = objectFound()
rects = ob.rectangle(img1.copy(), thresh, debug=True, mode='all')    
```

```python
maxrate = 0
roiF = None
for index, rect in enumerate(rects):
    #计算0-255比例, 0要尽可能多
    x,y,w,h = rect
    roi = thresh[y:y+h, x:x+w]
    rate = (np.sum(roi==0)/(roi.shape[0]*roi.shape[1])) + (0.1-index*0.01) #排名奖励
    if rate > maxrate:
        maxrate = rate
        roiF = roi

cv2.imshow("roi", roiF)
cv2.waitKey(0)
cv2.destroyAllWindows()
```

```python
print("计算mm/pixel : {},{},{}".format(20/roiF.shape[0], 20/roiF.shape[1], 0))
```

    计算mm/pixel : 0.024509803921568627,0.024009603841536616,0
    

## 角点检测
* 过滤出符合条件的角点，定义了一个筛选方法

```python
gray = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
blur = cv2.GaussianBlur(gray, (15,15), 0)
# _, thresh = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
_, thresh = cv2.threshold(blur, 150, 255, 0)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT,(12, 12))
erode = cv2.erode(thresh, kernel, 1)
mask = cv2.cornerHarris(erode, 40, 3, 0.04)
#result is dilated for marking the corners, not important
mask = cv2.dilate(mask,None)

# Threshold for an optimal value, it may vary depending on the image.
img11 = img1.copy()
img11[mask>0.01*mask.max()]=[0,0,255]
cv2.namedWindow("corner", cv2.WINDOW_NORMAL)
cv2.imshow("corner", img11)
cv2.imshow("erode", erode)
cv2.waitKey(0)
cv2.destroyAllWindows()
```

```python
gray1 = np.zeros_like(gray)
gray1[mask>0.01*mask.max()]=255
_, labels, stats, centroids = cv2.connectedComponentsWithStats(gray1)
cv2.namedWindow("labels", cv2.WINDOW_NORMAL)
cv2.imshow("labels", labels)
cv2.waitKey(0)
cv2.destroyAllWindows()
```

## 定义筛选函数

```python
def exitHelper(roi):
    """左右两个边孤岛和为3，上下同理"""
    h,w = roi.shape
    edge1 = roi[0:h-1, 0:10] #左
    edge2 = roi[0:h-1, w-10:w-1]#右
    edge3 = roi[0:10, 0:w-1] #上
    edge4 = roi[h-10:h-1, 0:w-1] #下
    connected255 = 0
    _, labels, stats, centroids1 = cv2.connectedComponentsWithStats(edge1)
    _, labels, stats, centroids2 = cv2.connectedComponentsWithStats(edge2)
    if len(centroids2) + len(centroids1) == 3:
        _, labels, stats, centroids1 = cv2.connectedComponentsWithStats(edge3)
        _, labels, stats, centroids2 = cv2.connectedComponentsWithStats(edge4)
        if len(centroids2) + len(centroids1) == 3:
            return True
    return False
    
wsize=100
result = []
for p in centroids:
    x,y = int(np.round(p[0])), int(np.round(p[1]))
    x1, y1, x2, y2 = x-wsize, y-wsize, x+wsize, y+wsize
    if x         x1 = 0
    if y         y1 = 0
    roi = thresh[y1:y2, x1:x2]
    if exitHelper(roi):
        result.append(p)
        cv2.imshow("roi", roi)
        cv2.waitKey(0)
        cv2.destroyAllWindows()
    
```

```python
#试算扫描线宽
wsize=100
step = 10 #pixel
data = []
for p in result:
    x,y = int(np.round(p[0])), int(np.round(p[1]))
    x1, y1, x2, y2 = x-wsize, y-wsize, x+wsize, y+wsize
    if x         x1 = 0
    if y         y1 = 0
    roi = thresh[y1:y2, x1:x2]
    h, w = roi.shape
    widthList = []
    #左右扫描
    cnts = []
    for i in range(0, w, step):
#         print(roi[0:h, i])
        if np.sum(roi[0:h, i])/255 < wsize and np.sum(roi[0:h, i])/255 > 0:
            cnts.append(np.sum(roi[0:h, i])/255)
    widthList.append([(x1+i, y1), cnts])
    #上下扫描
    cnts = []
    for i in range(0, h, step):
#         print(roi[i, 0:w])
        if np.sum(roi[i, 0:w])/255 < wsize and np.sum(roi[i, 0:w])/255 > 0:
            cnts.append(np.sum(roi[i, 0:w])/255)
    widthList.append([(x1, y1+i), cnts])
    data.append(widthList)
    
```

```python
img11 = img1.copy()
for line in data:
    for d in line:
        p, cnts = d
        cv2.circle(img11, p, 3, (0,255,0), 1)
        cv2.putText(img11,'{:.2f}'.format(np.median(cnts)*0.024), (p[0]+5,p[1]+5),cv2.FONT_HERSHEY_SIMPLEX,1,(0,0,255),2)
cv2.imwrite( "../day2.jpg", img11)
cv2.namedWindow("result", cv2.WINDOW_NORMAL)
cv2.imshow("result", img11)
cv2.waitKey(0)
cv2.destroyAllWindows()
```

```python
dataset={
    "lineA":{
        "width":[1.06,1.34,1.46,1.49,1.52,1.54,1.49,1.39,1.48,1.39,1.50,1.41,1.34,1.37,1.39,1.36,1.40,1.28,1.26]
    },
    "lineB":{
        "width":[1.15,1.24,1.19,1.25,1.33,1.29,1.36,1.17,1.31,1.33,1.37,1.32,1.37,1.44,1.37,1.4,1.43]
    }
}
```