【python-opencv】硬币检测-Toy模板网

这篇具有很好参考价值的文章主要介绍了【python-opencv】硬币检测。希望对大家有所帮助。如果存在错误或未考虑完全的地方，请大家不吝赐教，您也可以点击"举报违法"按钮提交疑问。

使用 python3.8.x，opencv

问题描述

使用图像处理技术，从照片中识别硬币的个数，并判断总价值。

设计思路1

使用简单特征识别

使用颜色特征，识别出5角硬币
使用半径大小，判断出1角和1元硬币。

具体操作

将图片转换为HSV颜色模型

hsv = cv2.cvtColor(imgROI, cv2.COLOR_BGR2HSV)
lowerYellowHSV = np.array([11,43,46])
upperYellowHSV = np.array([34,255,255])
mask=cv2.inRange(hsv,lowerb=lowerYellowHSV,upperb=upperYellowHSV)/255

图像识别硬币,图像处理,opencv,计算机视觉,python

部分代码

def coinSimpleDetection(img,circlesVector):
    # 简单识别，利用硬币的色彩和半径进行区分

    circlesNum = circlesVector.shape[0]
    # 将图片裁剪出来
    candidateImage = {}
    for i in range(circlesNum):
        information = {"radius":0,"value":0}
        col,row,radius = circlesVector[i,:]
        imgTemp = img.copy()
        imgROI = imgTemp[row-radius:row+radius,col-radius:col+radius,:]
        information["radius"] = radius

        # 识别5角硬币
        hsv = cv2.cvtColor(imgROI, cv2.COLOR_BGR2HSV)
        lowerYellowHSV = np.array([11,43,46])
        upperYellowHSV = np.array([34,255,255])
        mask=cv2.inRange(hsv,lowerb=lowerYellowHSV,upperb=upperYellowHSV)/255
        if np.sum(np.array(mask))/((row+2*radius)*(col+2*radius)) > 0.05:
            information["value"] = 0.5

        # 根据半径判断1块，新旧1角硬币
        if information["value"] == 0:
            if information["radius"] > 180 and information["radius"] < 250:
                information["value"] = 0.1
            if information["radius"] > 250 and information["radius"] < 300:
                information["value"] = 1.0
        candidateImage.update({i:information})
    cionsValue = np.sum([candidateImage[k]["value"] for k in range(circlesNum)])
    return candidateImage,cionsValue

设计思路2

模板匹配

提取轮廓信息 cv2.findContours()
拟合椭圆，提取ROI cv2.fitEllipse()
模板匹配cv2.matchTemplate()
选取最优匹配

源码

模板制作

有同学说不知道如何制作模板，其实，只需要将上面所说的模板匹配过程执行一半，获取到ROI之后，分别保存下来即可。
我把模板制作代码完整地贴出来吧

"""
模板制作
"""
import cv2
import numpy as np
import matplotlib.pyplot as plt

# 参数给定
numOfTimeToBlur = 3 # 去噪次数
blurKernelSize = 5 # 去噪核大小
cannyThreshold = 100 # canny 算子的低阈值
cannyThreshold2 = 2*cannyThreshold # canny 算子的高阈值，一般为低阈值的2~3倍
minAreaOfCircle = 800 # 最小圆形面积
numberOfTemplates = 8
threshold1=[50,50,50,50,20,30,40,30]
threshold2=[120,120,120,120,40,60,80,60]

# 图片路径
sourceImgPath = ".\\coin Image\\Test Image\\coin.jpg"
templateImgPath = ".\\coin Image\\Template Images\\"

if __name__ == "__main__":
    # 0.1 图像读取
    sourceImg = cv2.imread(sourceImgPath,cv2.IMREAD_COLOR)
    contourFilledWithColorImg = sourceImg.copy()

    # 1.1 转换为灰度图
    sourceImgGray = cv2.cvtColor(sourceImg,cv2.COLOR_BGR2GRAY)

    # 1.2 图像去噪
    for i in range(0,numOfTimeToBlur):
        sourceImgGray = cv2.GaussianBlur(sourceImgGray,(blurKernelSize,blurKernelSize),2.0,2.0)

    # 1.3 Canny 边缘检测
    sourceImgCanny = cv2.Canny(sourceImgGray,cannyThreshold,cannyThreshold2)

    # 1.4 膨胀canny图像
    dilationKernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(3,3),(1,1))
    cv2.dilate(sourceImgCanny,dilationKernel,sourceImgCanny)

    # 1.5 在边缘中找轮廓
    # CV_RETR_EXTERNAL只检测最外围轮廓
    # CV_CHAIN_APPROX_SIMPLE 仅保存轮廓的拐点信息
    contours, _ = cv2.findContours(sourceImgCanny,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)

    # 2 遍历每一个找到的轮廓
    matchingContours = []
    for currentContour in range(0,len(contours)):
        # 2.1 轮廓拐点小于5或者小于最小面积限制，跳过
        if len(contours[currentContour]) < 5 or cv2.contourArea(contours[currentContour])<= minAreaOfCircle:
            continue

        # 2.2 创建形状包围轮廓，得到轮廓的边界矩形的顶点和其相反点
        # boundingRectVals = cv2.boundingRect(contours[currentContour]) # x y w h
        boundingRect_x,boundingRect_y,boundingRect_w,boundingRect_h = cv2.boundingRect(contours[currentContour])
        p1 = (boundingRect_x,boundingRect_y)
        p2 = (boundingRect_x+boundingRect_w,boundingRect_y+boundingRect_h)
        bottom = (boundingRect_x,boundingRect_y + boundingRect_h + 50)

        # 2.3 求轮廓的外接椭圆
        ellipse = cv2.fitEllipse(contours[currentContour]) # [ (x, y) , (a, b), angle ] 椭圆中心位置，长短轴，旋转角度

        # 2.4 用红色绘制外接椭圆
        # 【这里进行绘制，会导致patch中有红色线，故将其移动到保存patch之后】
        # cv2.ellipse(sourceImg,ellipse,(0,0,255),3,cv2.LINE_AA)

        # 3 模板匹配
        # 3.1 创建与轮廓周围的边界矩形大小相同的patch
        patch = np.zeros((boundingRect_h,boundingRect_w,3),np.uint8)

        # 3.2 为轮廓区域填充颜色
        cv2.ellipse(contourFilledWithColorImg,ellipse,(111,222,111),cv2.FILLED,cv2.LINE_AA)

        # 3.3 创建填充的patch，用于从图像中提取硬币而忽略背景
        patchFilled = patch.copy()

        # 3.4 提取轮廓范围内的像素到patch
        sourceRow = boundingRect_y

        for patchRow in range(patch.shape[0]):
            sourceCol = boundingRect_x
            for patchCol in range(patch.shape[1]):
                patchFilled[patchRow,patchCol,:] = contourFilledWithColorImg[sourceRow,sourceCol,:]
                if patchFilled[patchRow,patchCol,0] == 111 and patchFilled[patchRow,patchCol,1] == 222 and patchFilled[patchRow,patchCol,2] == 111:
                    patch[patchRow,patchCol,:] = sourceImg[sourceRow,sourceCol,:]
                sourceCol += 1
            sourceRow += 1

        # 3.5 保存每一个patch为模板
        fpath = templateImgPath + "{}.jpg".format(currentContour)
        print(fpath)
        cv2.imwrite(fpath,patch)

        # 用红色绘制外接椭圆
        cv2.ellipse(sourceImg,ellipse,(0,0,255),3,cv2.LINE_AA)

    cv2.namedWindow("sourceImg",cv2.WINDOW_NORMAL|cv2.WINDOW_KEEPRATIO)
    cv2.imshow("sourceImg",sourceImg)
    cv2.waitKey(0)