OpenCV 特征匹配多个物体

Question

我如何在一张图像中找到多个同类型的物体。 我使用 ORB 特征点检测和暴力匹配器 (opencv = 3.2.0)。

我的源代码:

import numpy as np
import cv2
from matplotlib import pyplot as plt

MIN_MATCH_COUNT = 10

img1 = cv2.imread('box.png', 0)  # queryImage
img2 = cv2.imread('box1.png', 0) # trainImage

#img2 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)

# Initiate ORB detector
# 
orb = cv2.ORB_create(10000, 1.2, nlevels=9, edgeThreshold = 4)
#orb = cv2.ORB_create()

# find the keypoints and descriptors with SIFT
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)

FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
search_params = dict(checks = 50)

flann = cv2.FlannBasedMatcher(index_params, search_params)

des1 = np.float32(des1)
des2 = np.float32(des2)

# matches = flann.knnMatch(des1, des2, 2)

bf = cv2.BFMatcher()
matches = bf.knnMatch(des1, des2, k=2)

# store all the good matches as per Lowe's ratio test.
good = []
for m,n in matches:
    if m.distance < 0.7*n.distance:
        good.append(m)

if len(good)>3:
    src_pts = np.float32([ kp1[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
    dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good ]).reshape(-1,1,2)

    M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 2)

    if M is None:
        print ("No Homography")
    else:
        matchesMask = mask.ravel().tolist()

        h,w = img1.shape
        pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
        dst = cv2.perspectiveTransform(pts,M)

        img2 = cv2.polylines(img2,[np.int32(dst)],True,255,3, cv2.LINE_AA)

else:
    print ("Not enough matches are found - %d/%d" % (len(good),MIN_MATCH_COUNT))
    matchesMask = None

draw_params = dict(matchColor = (0,255,0), # draw matches in green color
                   singlePointColor = None,
                   matchesMask = matchesMask, # draw only inliers
                   flags = 2)

img3 = cv2.drawMatches(img1,kp1,img2,kp2,good,None,**draw_params)

plt.imshow(img3, 'gray'),plt.show()

但它只能找到一个查询图像实例。

查询图像

测试图像

结果

因此，它仅从两个图像中找到一个图像实例。 我做错了什么？

- V. Gai

---

找到第一个对象，计算变换，遮罩发现的对象区域，重复直到获得所有对象。 - Michał Gacka

@m3h0w 谢谢！ - V. Gai

是的，那听起来像个计划。 - Michał Gacka

现在没有时间阅读文档，但我认为可以合理地假设匹配算法正在寻找最佳匹配对象而不是多个对象。 - Michał Gacka

@V.Gai 您还可以查阅文献，了解处理这种情况的常见方法（严格特征匹配处理匹配描述符，没有对象假设）。Lowe 的 SIFT 论文提出了一种基于霍夫投票的方法。我最近发现了这篇论文：MAC-RANSAC：用于多个对象识别的鲁棒算法，但应该还有许多其他论文可供参考。 - Catree

Answer 1

我用 ORB 描述符寻找多个物体的源代码

import cv2
from matplotlib import pyplot as plt

MIN_MATCH_COUNT = 10

img1 = cv2.imread('box.png', 0)  # queryImage
img2 = cv2.imread('box1.png', 0) # trainImage

orb = cv2.ORB_create(10000, 1.2, nlevels=8, edgeThreshold = 5)

# find the keypoints and descriptors with ORB
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)

import numpy as np
from sklearn.cluster import MeanShift, estimate_bandwidth

x = np.array([kp2[0].pt])

for i in xrange(len(kp2)):
    x = np.append(x, [kp2[i].pt], axis=0)

x = x[1:len(x)]

bandwidth = estimate_bandwidth(x, quantile=0.1, n_samples=500)

ms = MeanShift(bandwidth=bandwidth, bin_seeding=True, cluster_all=True)
ms.fit(x)
labels = ms.labels_
cluster_centers = ms.cluster_centers_

labels_unique = np.unique(labels)
n_clusters_ = len(labels_unique)
print("number of estimated clusters : %d" % n_clusters_)

s = [None] * n_clusters_
for i in xrange(n_clusters_):
    l = ms.labels_
    d, = np.where(l == i)
    print(d.__len__())
    s[i] = list(kp2[xx] for xx in d)

des2_ = des2

for i in xrange(n_clusters_):

    kp2 = s[i]
    l = ms.labels_
    d, = np.where(l == i)
    des2 = des2_[d, ]

    FLANN_INDEX_KDTREE = 0
    index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
    search_params = dict(checks = 50)

    flann = cv2.FlannBasedMatcher(index_params, search_params)

    des1 = np.float32(des1)
    des2 = np.float32(des2)

    matches = flann.knnMatch(des1, des2, 2)

    # store all the good matches as per Lowe's ratio test.
    good = []
    for m,n in matches:
        if m.distance < 0.7*n.distance:
            good.append(m)

    if len(good)>3:
        src_pts = np.float32([ kp1[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
        dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good ]).reshape(-1,1,2)

        M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 2)

        if M is None:
            print ("No Homography")
        else:
            matchesMask = mask.ravel().tolist()

            h,w = img1.shape
            pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
            dst = cv2.perspectiveTransform(pts,M)

            img2 = cv2.polylines(img2,[np.int32(dst)],True,255,3, cv2.LINE_AA)

            draw_params = dict(matchColor=(0, 255, 0),  # draw matches in green color
                               singlePointColor=None,
                               matchesMask=matchesMask,  # draw only inliers
                               flags=2)

            img3 = cv2.drawMatches(img1, kp1, img2, kp2, good, None, **draw_params)

            plt.imshow(img3, 'gray'), plt.show()

    else:
        print ("Not enough matches are found - %d/%d" % (len(good),MIN_MATCH_COUNT))
        matchesMask = None

结果图片

- V. Gai

---

您的答案是所有搜索中最好的，但我需要C++实现。我找不到estimate_bandwidth和MeanShift的替代品，请问您能指点我吗？在C++中如何将找到的良好匹配分段成群？非常感谢！ - Suge

你好！我认为你可以在这里找到C++的MeanShift：https://docs.opencv.org/3.4.3/dc/d6b/group__video__track.html#ga432a563c94eaf179533ff1e83dbb65ea。关于聚类，可以参考这里：https://docs.opencv.org/2.4/modules/core/doc/clustering.html。 - V. Gai

MeanShift 和 Python 中的 scikit.learn 版本差别很大，它用于追踪摄像头中的对象，不能跟踪多个对象。而 OpenCV 中的 kmeans 需要先估计聚类数，这种情况让我感到无助。 - Suge

我在这个帖子中发布了我的尝试，您能否看一下？非常感谢，您的方法太完美了！https://dev59.com/Bek5XIcBkEYKwwoY7OHv - Suge

Answer 2

解决这个任务可以采用以下方法：

1. SIFT（SURF）+ MEAN SHIFT
2. Haar级联
3. HOG + SVM

- V. Gai