We Keep Coding

OpenCV: Reducing "fuzzy maxima" into single points

I have used:
result = cv2.matchTemplate(frame, template, cv2.TM_CCORR_NORMED)
to generate this output:
I need a list of (x, y) tuples at each of the local maxima (bright spots) in the result. Simply finding all points above a threshold doesn't work, since there are many such points around each maximum.
I can guarantee the minimum distance between any two maxima, which ought to help speed things up.
Is there an efficient technique for doing this?
(P.S.: this is cross-posted from https://forum.opencv.org/t/locating-local-maximums/1534)
update
Based on an excellent suggestion by Michael Lee, I've added skeletonizing to the thresholded image. It's close, but the skeletonizing still has many "worms" rather than single points. My processing flow is as follows:
# read the image
im = cv.imread("image.png", cv.IMREAD_GRAYSCALE)
# apply thresholding
ret, im2 = cv.threshold(im, args.threshold, 255, cv.THRESH_BINARY)
# dilate the thresholded image to eliminate "pinholes"
im3 = cv.dilate(im2, None, iterations=2)
# skeletonize the result
im4 = cv.ximgproc.thinning(im3, None, cv.ximgproc.THINNING_ZHANGSUEN)
# print the number of points found
x, y = np.nonzero(im5)
print(x.shape)
# => 1208
This is a step in the right direction, but there should be more like 220 points, not 1208.
Here are the intermediate results. As you can see in the last picture (skeletonized), there are still lots of little "worms" rather than single point. Is there a better approach?
Thresholded:
Dilated:
Skeletonized:

Update 2/14: Seems like skeletonization only took you part of the way there. Here's a better solution which I believe should get you the rest of the way. Here's how you would do it in scikit-image - maybe you can find the analog in OpenCV - seems like cv2.findContours would be a good start.
# mask is the thresholded image (before or after dilation should work, no skeletonization.
from skimage.measure import label, regionprops
labeled_image = label(mask)
output_points = [region.centroid for region in regionprops(labeled_image)]
Explanation: Label will convert your binary image into a labeled image, where each mask has a different integer value. Then, regionprops uses these labels in order to separate each mask, from which we can use the centroid property to compute the middle point from each - this is guaranteed to be a single point.
Simply finding all points above a threshold doesn't work, since there
are many such points around each maximum.
Actually, this does work - as long as you apply one more processing step. After thresholding, then we want to skeletonize. Scikit-image has a good function to achieve that here, which should give you a binary mask with single points.
Afterwards, you're probably going to want to run something like:
indices = zip(*np.where(skeleton))
to get your final points!

Based on Michael Lee's answer, here's the solution that worked for me (using all openCV rather than skimage):
# read in color image and create a grayscale copy
im = cv.imread("image.png")
img = cv.cvtColor(im, cv.COLOR_BGR2GRAY)
# apply thresholding
ret, im2 = cv.threshold(img, args.threshold, 255, cv.THRESH_BINARY)
# dilate the thresholded peaks to eliminate "pinholes"
im3 = cv.dilate(im2, None, iterations=2)
contours, hier = cv.findContours(im3, cv.RETR_TREE, cv.CHAIN_APPROX_SIMPLE)
print('found', len(contours), 'contours')
# draw a bounding box around each contour
for contour in contours:
x,y,w,h = cv.boundingRect(contour)
cv.rectangle(im, (x,y), (x+w,y+h), (255,0,0), 2)
cv.imshow('Contours', im)
cv.waitKey()
which results in just what we're looking for:

Detecting silver and reflecting balls with OpenCV

I am trying to detect silver balls reflecting the environment with OpenCV:
With black balls, I successfully did it by detecting circles:
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray,(5,5),0);
gray = cv2.medianBlur(gray,5)
gray = cv2.adaptiveThreshold(gray,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY,11,3.5)
kernel = np.ones((3,3),np.uint8)
gray = cv2.erode(gray,kernel,iterations = 1)
gray = cv2.dilate(gray,kernel,iterations = 1)
circles = cv2.HoughCircles(gray, cv.CV_HOUGH_GRADIENT, 1, 260, \
param1=30, param2=65, minRadius=0, maxRadius=0)
But when using the program with silver balls, we don't get any result.
When looking at the edges calculated by the program, the edges of the ball are quite sharp. But the Code is not recognizing any ball.
How do I improve the detection rate of the silver ball? I think of two ways doing that:
- Improving edge calculation
- Make the circle detection accept an image with unclear edges
Is that possible? What is the best way of doing so?
Help is very appreciated.

You have to tune your parameters. The HoughCircles function does a good job in detecting circles (even with gaps). Note that HoughCircles performs an internal binarization using canny edge detection. Thus, you don't have to do thresholding.
Given your image above, the code
import cv2
from matplotlib import pyplot as plt
import numpy as np
PATH = 'path/to/the/image.jpg'
img = cv2.imread(PATH, cv2.IMREAD_GRAYSCALE)
plt.imshow(img, cmap='gray')
circles = cv2.HoughCircles(img, cv2.HOUGH_GRADIENT, 1, 20, param1=130, param2=30, minRadius=0, maxRadius=0)
if circles is not None:
for x, y, r in circles[0]:
c = plt.Circle((x, y), r, fill=False, lw=3, ec='C1')
plt.gca().add_patch(c)
plt.gcf().set_size_inches((12, 8))
plt.show()
yields the result
What do the different parameters mean?
The function signature is defined as
cv.HoughCircles(image, method, dp, minDist[, circles[, param1[, param2[, minRadius[, maxRadius]]]]])
image and circles are self explanatory and will be skipped.
method
Specifies the variant of the hough algorithm that is used internally. As stated in the documentation, only HOUGH_GRADIENT is support atm. This method utilizes the 21HT (p.2, THE 2-1 HOUGH TRANSFORM) algorithm. The major advantage of this variant lies in the reduction of memory usage. The standard way of detecting circles using hough transform requires a search in a 3D hough space (x, y and radius). However, with 21HT your hough space is reduced to only 2 dimensions, which lowers the memory consumption by a fair amount.
dp
The dp parameter sets the inverse accumulator resolution. A good explanation can be found here. Note that this explanation uses the standard hough transform as an example. But the effect is the same for 21HT. The accumulator for 21HT is just a bit different as for the standard HT.
minDist
Simply specifies the minimum distance between the centers of circles. In the code example above it's set to 20 which means that the centers of two detected circles have to be at least 20 pixels away from each other. I'm not sure how opencv filters the circles out, but scanning the source code it looks like circles with lower matches are thrown out.
param1
Specifies the thresholds that are passed to the Canny Edge algorithm. Basically it's called like cv2.Canny(image, param1 / 2, param1).
param2
This paragraph should probably be validated by someone who is more familiar with the opencv source code.
param2 specifies the accumulator threshold. This value decides how complete a circle has to be in order to count as a valid circle. I'm not sure in which unit the parameter is given, tho. But (scanning the source code again) it looks like it is an absolute vote threshold (meaning that it is directly affected by different radii).
The image below shows different circles (or what can be identified as a circle). The further you go to the right, the lower the threshold has to be in order to detect that circle.
minRadius and maxRadius
Simply limits the circle search in the range [minRadius, maxRadius] for the radius. This is useful (and can increase performance) if you can approximate (or know) the size of the circles you are searching for.

OpenCV Hough Circle Transform Not Working

I have followed OpenCV's tutorial here for circle detection on my Raspberry Pi. This is the code that I am using which is the same as the tutorial except a different image.
import cv2
import numpy as np
img = cv2.imread('watch.jpg',0)
img = cv2.medianBlur(img,5)
cimg = cv2.cvtColor(img,cv2.COLOR_GRAY2BGR)
circles = cv2.HoughCircles(img,cv2.HOUGH_GRADIENT,1,20,
param1=50,param2=30,minRadius=0,maxRadius=0)
circles = np.uint16(np.around(circles))
for i in circles[0,:]:
cv2.circle(cimg,(i[0],i[1]),i[2],(0,255,0),2)
cv2.circle(cimg,(i[0],i[1]),2,(0,0,255),3)
cv2.imshow('image',cimg)
cv2.waitKey(0)
cv2.destroyAllWindows()
Then when I ran the script this is what I was presented with this
and this is the original image
What is causing this to happen?
Thank You in Advance!
Edit:

The large amount of circles generated by Hough Circle Transform is caused by the low value of the threshold for center detection, which is param2 in cv2.HoughCircles in your case.
So try to increase the value of param2 to avoid false detections.
Also you can adjust minRadius and maxRadius values for better results.
EDIT:
I have just tried example from here and changed only param2 to 10, minRadius to 30 and maxRadius to 50. The result is good enough:
The example from the link above is written with C++, but you can compare parameters and the sequence of functions invocations to refine your own algorithm.

Hough transform not detecting correct circle

I'm working on a project in which I have to detect Traffic lights (circles obviously). Now I am working with a sample image I picked up from a spot, however after all my efforts I can't get the code to detect the proper circle(light).
Here is the code:-
# import the necessary packages
import numpy as np
import cv2
image = cv2.imread('circleTestsmall.png')
output = image.copy()
# Apply Guassian Blur to smooth the image
blur = cv2.GaussianBlur(image,(9,9),0)
gray = cv2.cvtColor(blur, cv2.COLOR_BGR2GRAY)
# detect circles in the image
circles = cv2.HoughCircles(gray, cv2.HOUGH_GRADIENT, 1.2, 200)
# ensure at least some circles were found
if circles is not None:
# convert the (x, y) coordinates and radius of the circles to integers
circles = np.round(circles[0, :]).astype("int")
# loop over the (x, y) coordinates and radius of the circles
for (x, y, r) in circles:
# draw the circle in the output image, then draw a rectangle
# corresponding to the center of the circle
cv2.circle(output, (x, y), r, (0, 255, 0), 4)
cv2.rectangle(output, (x - 5, y - 5), (x + 5, y + 5), (0, 128, 255), -1)
# show the output image
cv2.imshow("output", output)
cv2.imshow('Blur', blur)
cv2.waitKey(0)
The image in which I want to detect the circle-
This is what the output image is:-
I tried playing with the Gaussian blur radius values and the minDist parameter in hough transform but didn't get much of success.
Can anybody point me in the right direction?
P.S- Some out of topic questions but crucial ones to my project-
1. My computer takes about 6-7 seconds to show the final image. Is my code bad or my computer is? My specs are - Intel i3 M350 2.6 GHz(first gen), 6GB RAM, Intel HD Graphics 1000 1625 MB.
2. Will the hough transform work on a binary thresholded image directly?
3. Will this code run fast enough on a Raspberry Pi 3 to be realtime? (I gotta mount it on a moving autonomous robot.)
Thank you!

First of all you should restrict your parameters a bit.
Please refer to: http://docs.opencv.org/2.4/modules/imgproc/doc/feature_detection.html#houghcircles
At least set reasonable values for min and max radius. Try to find that one particular circle first. If you succeed increase your radius tolerance.
Hough transform is a brute force method. It will try any possible radius for every edge pixel in the image. That's why it is not very suitable for real time applications. Especially if you do not provide proper parameters and input. You have no radius limits atm. So you will calculate hundreds, if not thousands of circles for every pixel...
In your case the trafficlight also is not very round so the accumulated result won't be very good. Try finding highly saturated, bright, compact blobs of a reasonable size instead. It should be faster and more robust.
You can further reduce processing time if you restrict the image size. I guess you can assume that the traffic light will always be in the upper half of your image. So omit the lower half. Traffic lights will always be green, red or yellow. Remove everything that is not of that color... I think you get what I mean...

I think that you should first perform a color segmentation based on the stoplight colors. It will tremendously reduce the ROI. Then you can apply the Hough Transform on the ROI edges only (because you want the contour).

Another restriction: Only accept circles where the inside color is homogenous.This would throw out all the false hits in the above example.

Circular Hough Transform misses circles

I've read a lot about the Circular Hough transform on Stack Overflow, but I seem to be missing something. I wrote a program that is supposed to detect the circles of a "Bull's Eye" target. However, even after playing with the parameters, the algorithm is quite bad - it ignores most of the circles and one time it finds a circle but seems to "wander off". I've even tried applying an "Unsharp Mask" to no avail. I have added my code, the image I started with and the output. I hope someone can point me at the right direction.
import cv2
import cv2.cv as cv
import numpy as np
import math
# Load Image
img = cv2.imread('circles1.png',0)
# Apply Unsharp Mask
tmp = cv2.medianBlur(img,5)
img = cv2.addWeighted(img,1.5,tmp,-0.5,0)
cimg = cv2.cvtColor(img,cv2.COLOR_GRAY2BGR)
# Hough Transform
circles = cv2.HoughCircles(img,cv.CV_HOUGH_GRADIENT,1,5,
param1=100,param2=100,minRadius=0,maxRadius=0)
circles = np.uint16(np.around(circles))
# Go over circles, eliminating the ones that are not cocentric enough
height, width = img.shape
center = (width/2,height/2)
for i in circles[0,:]:
# draw the outer circle
if math.sqrt((center[0]-i[0])**2 + (center[1]-i[1])**2) < 15:
cv2.circle(cimg,(i[0],i[1]),i[2],(0,255,0),1)
# draw the center of the circle
cv2.circle(cimg,(i[0],i[1]),2,(0,0,255),3)
cv2.imshow('detected circles',cimg)
cv2.waitKey(0)
cv2.destroyAllWindows()
Quick explanation: I load the image, apply Unsharp Mask, use the Hough Transfrom to detect circles, then draw the circles that are close to the center (I found that the other circles are false circles).
I tried playing with the parameters, and this is the best I got. I feel like this is a simple enough problem which has me buffled. I appriciate any help.
My input image:
My output image:

As I mentioned in my comment, you'll need to run successive iterations of cv2.HoughCircles for different range of radii to ensure that you get all of the circles. With the way the Circular Hough Transform works, specifying a minimum and maximum radius that has quite a large range will be inaccurate and will also be slow. They don't tell you this in the documentation, but for the Circular Hough Transform to work successfully, the following two things need to be valid:
maxRadius < 3*minRadius
maxRadius - minRadius < 100
With the above, blindly making the minimum radius very small and the maximum radius very large won't give you great results. Therefore, what you could do is start at... say...radius=1, then iterate up to radius=300 in steps of 20. Between each chunk of 20, run cv2.HoughCircles and update your image with these contours.
Doing this requires very little modification to your code. BTW, I removed the unsharp masking it because I was getting poor results with it. I also changed a couple of parameters in cv2.HoughCircles slightly to get this to work as best as possible given your situation:
import cv2
import cv2.cv as cv
import numpy as np
import math
# Load Image
img = cv2.imread('circles1.png',0)
cimg = cv2.cvtColor(img,cv2.COLOR_GRAY2BGR)
# Specify different radii
radii = np.arange(0,310,10)
# For each pair of radii...
for idx in range(len(radii)-1):
# Get the minimum and maximum radius
# Note you need to add 1 to each minimum
# as the maximum of the previous pair covers this new minimum
minRadius = radii[idx]+1
maxRadius = radii[idx+1]
# Hough Transform - Change here
circles = cv2.HoughCircles(img,cv.CV_HOUGH_GRADIENT,1,5,
param1=25,param2=75,minRadius=minRadius,maxRadius=maxRadius)
# Skip if no circles are detected - Change here
if circles is None:
continue
circles = np.uint16(np.around(circles))
# Go over circles, eliminating the ones that are not cocentric enough
height, width = img.shape
center = (width/2,height/2)
for i in circles[0,:]:
# draw the outer circle
if math.sqrt((center[0]-i[0])**2 + (center[1]-i[1])**2) < 15:
cv2.circle(cimg,(i[0],i[1]),i[2],(0,255,0),1)
# draw the center of the circle
cv2.circle(cimg,(i[0],i[1]),2,(0,0,255),3)
cv2.imshow('detected circles',cimg)
cv2.waitKey(0)
cv2.destroyAllWindows()
I get this figure:
Unfortunately it isn't perfect as it doesn't detect all of the circles. You'll have to play around with the cv2.HoughCircles function until you get good results.
However, I wouldn't recommend using cv2.HoughCircles here. May I suggest using cv2.findContours instead? This finds all of the contours in the image. In this case, these will be the black circles. However, you need to reverse the image because cv2.findContours assumes non-zero pixels are object pixels, so we can subtract 255 from the image assuming a np.uint8 type:
# Make copy of original image
cimg2 = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
# Find contours
contours,_ = cv2.findContours(255 - img, cv2.RETR_LIST, cv.CV_CHAIN_APPROX_NONE)
# Draw all detected contours on image in green with a thickness of 1 pixel
cv2.drawContours(cimg2, contours, -1, color=(0,255,0), thickness=1)
# Show the image
cv2.imshow('detected circles', cimg2)
cv2.waitKey(0)
cv2.destroyAllWindows()
This is what I get: