OpenCV Tutorials

We can play video files directly from the worldwide web with OpenCV using the cv2.VideoCapture() function. OpenCV does not have default sound support as it is a computer vision library. You can manipulate the frames directly from the stream, but if the frame manipulation is too computationally expensive the video will lag.

#! /usr/bin/python

import cv2

vc = cv2.VideoCapture('https://archive.org/download/bliptv-20131014-102441-Brigittedale-WhatToTalkAbout167/bliptv-20131014-102441-Brigittedale-WhatToTalkAbout167.mp4')
c=1
fps = 24

if vc.isOpened():
    rval , frame = vc.read()
else:
    rval = False

while rval:
    rval, frame = vc.read()
    cv2.imshow("Result",frame)
    cv2.waitKey(1000 / fps);
vc.release()

#! /usr/bin/python

import cv2

vc = cv2.VideoCapture('https://archive.org/download/bliptv-20131014-102441-Brigittedale-WhatToTalkAbout167/bliptv-20131014-102441-Brigittedale-WhatToTalkAbout167.mp4')

c=1

fps = 24

if vc.isOpened():

rval , frame = vc.read()

else:

rval = False

while rval:

rval, frame = vc.read()

cv2.imshow("Result",frame)

cv2.waitKey(1000 / fps);

vc.release()

Other than template matching, there are other algorithms to find objects in images. A group of algorithms based on keypoints is an example of that. Every keypoint must be a unique point in terms of contrast, edges etc. Once these keypoints are extracted the program can match them with keypoints of other images.

Let us start by extracting keypoints from an input image. To do this, we use the cv2.FeatureDetector_create() method. As input it takes the name of a keypoint detection algorithm. There is a large variety of this: MSER, SURF, ORB and many more.

import cv2
img = cv2.imread('house1.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE);

fd = cv2.FeatureDetector_create('ORB')
kpts = fd.detect(img)

for k in kpts:
  x,y = k.pt
  x = int(x)
  y = int(y)
  cv2.rectangle(img,(x,y), (x+2,y+2),(0,0,255),2)

cv2.imshow("Result",img)
cv2.waitKey(0);

import cv2

img = cv2.imread('house1.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE);

fd = cv2.FeatureDetector_create('ORB')

kpts = fd.detect(img)

for k in kpts:

x,y = k.pt

x = int(x)

y = int(y)

cv2.rectangle(img,(x,y), (x+2,y+2),(0,0,255),2)

cv2.imshow("Result",img)

cv2.waitKey(0);

Output:

The next step, is to take the keypoints from another image and match them. We can not simply match 2d vectors (x,y). Every keypoint will be described using a descriptor or patch. Finally, we need a matcher object to match all these patches.

import cv2

img = cv2.imread('house1.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE);
img2 = cv2.imread('house2.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE);

fd = cv2.FeatureDetector_create('ORB')
kpts = fd.detect(img)
kpts2 = fd.detect(img2)

# Now that we have the keypoints we must describe these points (x,y)
# and match them.
descriptor = cv2.DescriptorExtractor_create("BRIEF")
matcher = cv2.DescriptorMatcher_create("BruteForce-Hamming")


# descriptors (we must describe the points in some way)
k1, d1 = descriptor.compute(img, kpts)
k2, d2 = descriptor.compute(img2, kpts2)

# match the keypoints
matches = matcher.match(d1, d2)

# similarity
print '#matches:', len(matches)

import cv2

img = cv2.imread('house1.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE);

img2 = cv2.imread('house2.jpg', cv2.CV_LOAD_IMAGE_GRAYSCALE);

fd = cv2.FeatureDetector_create('ORB')

kpts = fd.detect(img)

kpts2 = fd.detect(img2)

# Now that we have the keypoints we must describe these points (x,y)

# and match them.

descriptor = cv2.DescriptorExtractor_create("BRIEF")

matcher = cv2.DescriptorMatcher_create("BruteForce-Hamming")

# descriptors (we must describe the points in some way)

k1, d1 = descriptor.compute(img, kpts)

k2, d2 = descriptor.compute(img2, kpts2)

# match the keypoints

matches = matcher.match(d1, d2)

# similarity

print '#matches:', len(matches)

The program will output the number of matching keypoints. This is the similarity between image1 and image2. The similarity will rise with the number of matching keypoints found.

Output:

$ python matcher.py
#matches: 420

1 2	$ python matcher.py #matches: 420

OpenCV play video from the web

Image similarity with keypoints in Python