绝对不是最佳的（三个循环），但这是有效的。
如果你有很多盒子，它应该被优化。
它基本上检查每一个项目，并将它们与已经记录的盒子进行比较。
如果当前框与组中的一个框有交集，则将其添加到该组。

import cv2 as cv
import numpy as np

img = cv.imread('./input.jpg')
image = cv.resize(img, None, None, 0.4, 0.4)

gray = cv.cvtColor(image, cv.COLOR_BGR2GRAY)
_, thresh = cv.threshold(gray, 190, 255, cv.THRESH_BINARY_INV)
kernel = np.ones((1, 1), np.uint8)
img_dilation = cv.dilate(thresh, kernel, iterations=1)
ctrs, _ = cv.findContours(img_dilation.copy(), cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
sorted_ctrs = sorted(ctrs, key=lambda ctr: cv.boundingRect(ctr)[0])
bounding_rectangles = []

for i, _ctr in enumerate(sorted_ctrs):
    x, y, w, h = cv.boundingRect(_ctr)
    bounding_rectangles.append([x, y, w, h])
    _roi = image[y:y + h, x:x + w]
    cv.rectangle(image, (x, y), (x + w, y + h), (36, 255, 12), 2)
    
    if w > 2 and h > 2:
        roi = cv.resize(_roi, (224, 224))
        
cv.imshow("Output", image)
cv.waitKey(0)

groups = [[bounding_rectangles[0]]]
added_index = [0]

for i, rect_item in enumerate(bounding_rectangles[1:]):
    group_found = False
    
    for j, group_items in enumerate(groups):
        
        for k, group_item in enumerate(group_items):
            if (group_item[0] <= rect_item[0] <= group_item[0] + group_item[2]) and i + 1 not in added_index:
                added_index.append(i + 1)
                groups[j].append(rect_item)
                group_found = True
                break
                
        if group_found is True:
            break
                
    if group_found is False:
        added_index.append(i + 1)
        groups.append([rect_item])

print(f"Groups found: {len(groups)}")

for i, item in enumerate(groups):
    print(f"Group {i + 1}: Box quantity = {len(item)}")