需要使用opencv识别“托盘是空的还是不空的”

Question

我想在python中使用OpenCV识别给定图像中的托盘是否为空。

以下是我尝试过的

1. 使用下面的代码检测最大的矩形并裁剪

kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3,3))
thresh = cv2.erode(thresh, kernel, iterations=4)
thresh = cv2.dilate(thresh, kernel, iterations=4)
cnts, hier = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

# Find the biggest contour
max_area = -1
max_c = 0
for i in range(len(cnts)):
    contour = cnts[i]
    area = cv2.contourArea(contour)
    if (area > max_area):
        max_area = area
        max_c = i
contour = cnts[max_c]

areas = [cv2.contourArea(c) for c in cnts]
max_index = np.argmax(areas)
cnt=cnts[max_index]

x,y,w,h = cv2.boundingRect(cnt)

crop_img = img[y:y+h, x:x+w]

#Create empty mask and flood fill
mask = np.zeros(edges.shape)
#Create 3-channel alpha mask
mask_stack = np.dstack([mask]*3)

它适用于中型和大型物体，但当我放硬币等小型物体时，这种方法不起作用，因为托盘也有一些划痕和灰尘。这是我在python中使用OpenCV的第一个项目

请帮助找到满足此要求的解决方案。

Answer 1

我建议您：

< li >进行相机校准，使图像中的线条变直，以及 < li >将您的相机与生产线对齐，这样您可以简化图像处理并使其更加稳定，以及 < li >尽可能使用更好的照明条件。

鉴于上面的图片，这里有一个蛮力解决方案：

• 您可以首先通过HSV分割提取感兴趣区域（ROI），然后进行轮廓检测，然后使用此SO答案提取托盘的矩形形状

#!/usr/bin/python3
# -*- coding: utf-8 -*-

import cv2
import numpy as np

# Ref: https://stackoverflow.com/questions/37177811
def crop_minAreaRect(img, rect):

    box = cv2.boxPoints(rect) 
    box = np.int0(box)

    W = rect[1][0]
    H = rect[1][1]

    Xs = [i[0] for i in box]
    Ys = [i[1] for i in box]
    x1 = min(Xs)
    x2 = max(Xs)
    y1 = min(Ys)
    y2 = max(Ys)

    angle = rect[2]
    
    rotated = False

    if angle < -45:
        angle += 90
        rotated = True
    # Center of rectangle in source image
    center = ((x1+x2)/2,(y1+y2)/2)
    # Size of the upright rectangle bounding the rotated rectangle
    size = (x2-x1, y2-y1)
    M = cv2.getRotationMatrix2D((size[0]/2, size[1]/2), angle, 1.0)
    # Cropped upright rectangle
    cropped = cv2.getRectSubPix(img, size, center)
    cropped = cv2.warpAffine(cropped, M, size)
    croppedW = W if not rotated else H 
    croppedH = H if not rotated else W
    # Final cropped & rotated rectangle
    croppedRotated = cv2.getRectSubPix(cropped, (int(croppedW),int(croppedH)), (size[0]/2, size[1]/2))
    return croppedRotated

def ROI(img):
    hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)

    lower_hsv = np.array([0, 0, 135]) # 0, 0, 135
    higher_hsv = np.array([180, 20, 240]) # 180, 20, 240

    mask = cv2.inRange(hsv, lower_hsv, higher_hsv)

    seg = cv2.bitwise_and(img, img, mask=mask)

    seg_gray = cv2.cvtColor(seg, cv2.COLOR_BGR2GRAY)

    k1 = 51
    seg_gauss = cv2.GaussianBlur(seg_gray, (k1, k1), 0)
    seg_th = cv2.threshold(seg_gauss, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]

    contours, hierarchy = cv2.findContours(image=seg_th, mode=cv2.RETR_EXTERNAL, method=cv2.CHAIN_APPROX_NONE)
    for i, cnt in enumerate(contours):
        if  cv2.contourArea(cnt)>1000000:
            x,y,w,h = cv2.boundingRect(cnt)     
            rect = cv2.minAreaRect(cnt)

    roi = crop_minAreaRect(img, rect)
    return roi

• 当您将问题简化为特定的ROI时，您可以使用边缘检测、形态学和过滤，如下所示：

def ImProc(roi):

    gray = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)

    k_smooth = 5
    gauss = cv2.GaussianBlur(gray, (k_smooth, k_smooth), 0)

    # # Canny
    edges = cv2.Canny(gauss,0,255)

    k = 31 #300
    kernel = np.ones((k,k),np.uint8)
    mph = cv2.morphologyEx(edges, cv2.MORPH_CLOSE, kernel)

    res = cv2.medianBlur(mph, 21)

    return res

我将这种方法应用于给定的图像：

empty = cv2.imread("empty.jpg")
full = cv2.imread("full.jpg")

roi_empty = ROI(empty)
roi_full = ROI(full)

res_empty = ImProc(roi_empty)
res_full = ImProc(roi_full)

cv2.namedWindow("res_full", cv2.WINDOW_NORMAL)
cv2.imshow("res_full", res_full)

cv2.waitKey(0)

对于满托盘（旋转），我得到了以下结果：

请注意，此解决方案受不同参数的影响，为了更好地控制它们并获得稳定的结果，我建议您考虑上面的建议。