Question

我正在尝试使用SURF和kNN对对象进行分类。该代码运行良好，但它偶尔会崩溃并显示“分段错误”。我不确定我是否做错了但是我很确定它已经纠正了。如果您想重现问题，这是输入文件。

import numpy as np
import cv2
import sys

trainfile = ['/home/nuntipat/Documents/Dataset/Bank/Training/15_20_front.jpg'
            , '/home/nuntipat/Documents/Dataset/Bank/Training/15_50_front.jpg'
            , '/home/nuntipat/Documents/Dataset/Bank/Training/15_100_front.jpg'
            , '/home/nuntipat/Documents/Dataset/Bank/Training/15_500_front.jpg'
            , '/home/nuntipat/Documents/Dataset/Bank/Training/15_1000_front.jpg'
            , '/home/nuntipat/Documents/Dataset/Bank/Training/16_20_front.jpg'
            , '/home/nuntipat/Documents/Dataset/Bank/Training/16_50_front.jpg'
            , '/home/nuntipat/Documents/Dataset/Bank/Training/16_500_front.jpg']
testfile = '/home/nuntipat/Documents/Dataset/Bank/20_1.jpg'

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

# Initiate FLANN matcher
flann = cv2.FlannBasedMatcher(index_params, search_params)

# Initiate SURF detector
surf = cv2.xfeatures2d.SURF_create(500)

# Create list of describtor
descriptor = []
for file in trainfile:
    img = cv2.imread(file)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    kp, des = surf.detectAndCompute(gray, None)
    descriptor.append(des)

# Clasify using test file
img = cv2.imread(testfile)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
kp1, des = surf.detectAndCompute(gray, None)

maxCount = 0
for i, d in enumerate(descriptor):  
    matches = flann.knnMatch(d, des, k=2)

    count = 0

    # ratio test as per Lowe's paper
    for (m,n) in matches:
        if m.distance < 0.7 * n.distance:
            count += 1

    if count > maxCount:
        maxCount = count
        maxMatch = i

print maxMatch

在我编写此代码之前，我尝试创建一个包含每个训练数据的kNN模型，并且只进行一次匹配。但它始终失败并导致“flann.add（descriptors）”处的分段错误。

import numpy as np
import cv2

trainfile = ['/home/nuntipat/Documents/Dataset/Bank/100_1.jpg', '/home/nuntipat/Documents/Dataset/Bank/100_2.jpg', '/home/nuntipat/Documents/Dataset/Bank/100_3.jpg']
testfile = '/home/nuntipat/Documents/Dataset/Bank/100_1.jpg'

# FLANN parameters
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
search_params = dict(checks=50)   # or pass empty dictionary

# Initiate FLANN matcher
flann = cv2.FlannBasedMatcher(index_params, search_params)

# Initiate SURF detector
surf = cv2.xfeatures2d.SURF_create()

# Train FLANN
for file in trainfile:
    img = cv2.imread(file)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

    keypoints, descriptors = surf.detectAndCompute(gray, None)

    flann.add(descriptors)

非常感谢你的帮助。

Answer 1

在失败时，可能会出现空白图像或描述符很少的图像。然后描述符矩阵为空，因此失败。

Answer 2

在此链接中说明如下：

flann.add([descriptors])

http://answers.opencv.org/question/44592/flann-index-in-python-training-fails-with-segfault/

希望它有所帮助！

OpenCV Python：使用FlannBasedMatcher时偶尔会出现分段错误

2 个答案: