dastaan90's profile - activity

This issue was raised on and solved. (link : http://code.opencv.org/issues/4438)

You can always compile an older branch from the source.

I ended up using histogram comparison for thresolded image for my defected pieces. System has to be calibrated earlier for "good" pieces and any defect will result in less histogram values. If it's below a certain thresold, it is declared as a rejected piece.

Angle of camera and piece position are very significant in this use case as certain variation will lead to totally different data.

Hope it helps.

I ended up using DC light source with a kind of diffusion paper with my entire setup covered so that natural light doesn't affect it anymore. It works so reliable with this simple idea. It can help in any light sensitive image processing application.

Hope it helps someone who stumbles across this. :)

It might be because it is searching for those libraires in native environment and somehow failing to get it. An ideal way should be to build 3rd party libraries on the go by giving cmake parameter -D BUI:D_<lib>=ON. I'm not sure if it includes zlib. But i have done it with TBB and it helped me. Hope it helps.

Can you tell us what you're using ? Haar / LBP ?

I am working on a project where I need to detect color of a particular region of the object. While It's very easy to get RGB value from the frame of a particular object. Results fluctuate a lot when lighting changes. Ie, from natural light to artificial ones. I am wondering if there is anything to eliminate minor lighting changes to get a reliable color detection.

Things I have tried so far.

1. Tried converting RGB to YCrCy and try hist_equalization on Y part.
2. Converted from BGR to gray, hist_equalizarion and converted back to RGB.

Any suggestions ? Thanks in advance.

If you're having a linux box, use "ssh -X ..". If using putty, there is a seperate option for X11 forwarding. You will have to enable it from there. Hope it helps.

How are you invoking ./raspicamtest ? If you're invoking it over ssh connection, you may need to enable X11 forwarding in your ssh connection. You can't invoke a GUI application over simple SSH connection.

@theodore Not necessarily. Sometimes, defects are so minor / major that convexity defects can not be used to measure it. I will try to add some more images just to give you more hints on defect types.

Hello, I'm working on a project where I need to find minor defects and report if it is defected.

Here are pictures of good and bad samples. Please note that these images are a result of post thresolding stage. Ie, noice removal and all pre-processing has been done. Gray outline that you see is convexHull curve for the contour (object).

Good sample

Bad sample (defect at top left corner)

Things I've tried till now.

1. Histogram matching of thresolded images.
2. Contour area and minAreaRect area comparing.

However, each of this method works on a particular defect but not all.

Any suggestions / tips for better results ?

It reminds me of one of my project while I was studying. First let me ask you something. What do you mean by direction ? If you meant iris position, you can detect eye and take iris template at the start of the program and later do template matching in your region of interest and keep track of the iris. Hope it helps.

Can you post your program ? So that ppl can help you in debuggin. To me it looks like, image is not getting stored in img.

It shouldn't be very difficult. You should have different paths while "cmaking" it. For example, you should specify different install paths using -DCMAKE_INSTALL_PREFIX= flag.

For example,

for opencv 2.4.9,

-DCMAKE_INSTALL_PREFIX=/path/to/opencv2.4.9/

while

for opencv 3.0.0

-DCMAKE_INSTALL_PREFIX=/path/to/opencv3.0.0

Do not confuse this path with source path.

And then real "magic" is in the makefile to compile your own code. Linking path and include paths have to be specified as per the opencv version you want.

For example : You have opencv 2.4.9 installed in /path/to/opencv2.4.9/lib/ and opencv 3.0.0 install in /path/to/opencv3.0.0/lib/ as per mentioned in that link, you will have makefile as per below:

Makefile for opencv 2.4.9

 CPP = g++

# OpenCV 2.4.9
CPPFLAGS = -L/path/to/opencv2.4.9/lib \
       -I/path/to/opencv2.4.9/include

all: test

test: main.cpp
    $(CPP) $(CPPFLAGS) $^ -o $@

and Makefile for opencv 3.0.0

 CPP = g++

# OpenCV 2.4.9
CPPFLAGS = -L/path/to/opencv3.0.0/lib \
       -I/path/to/opencv3.0.0/include

all: test

test: main.cpp
    $(CPP) $(CPPFLAGS) $^ -o $@

You might be confused as in that link, there are two CPPFLAGS defined in the same file.

Hope it clears your doubt.

Cool. Looks like you didn't have ffmpeg path set.

Can you post your cmake parameters as well ?

Did you look at this link ?

I am not sure if this would work for you but you can make use of contourArea method to get contour area and filter out some small contours on a basis of that. Here is the link.. Example code as below:

threshold_area = 10000     #threshold area 
contours, hierarchy = cv2.findContours(threshold,cv2.RETR_TREE,cv2.CHAIN_APPROX_NONE)   
for cnt in contours:        
    area = cv2.contourArea(cnt)         
    if area > threshold_area:                   
         #Put your code in here

Another elegant solution would be get a minimum area rect and get width and height from rect's 4 coordinates (by calculating distance on your own). But I'm not sure if you're using it to detect regular shaped contours and not.

And as far as I understand, you code should be rearranged as follow for it to work:

contours, hierarchy = cv2.findContours(im,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
for cnt in contours
    rect = cv2.minAreaRect(cnt)       #I have used min Area rect for better result
    width = rect[1][0]
    height = rect[1][1]
    if (width<widthmax) and (height <heightmax) and (width >= widthMin) and (height > heightMin):
        #put your code here if desired contour is detected

Hope it helps.

@ryanvade Let us know if it worked or not.

Can you tell us what you have tried so far by posting your code ?

There could be two possibilities.

1. You don't have IPP installed and trying to compile opencv with IPP.
2. You don't have IPP installed properly ie, paths are not properly installed.

Possible solution.

1. Make sure IPP is installed properly and run basis IPP tests if necessary.
2. If it's not installed, you can add this flag -D BUILD_IPP=ON to build library with opencv compilation.

Hope it helps.

@LBerger You're correct, ie you can make use of already installed libraries (like you mentioned with dll) or you can build a 3rd party library from source at the time of building OpenCV as per I understand and BUILD_TBB is for later option, ie, building on the go as you compile OpenCV.

Try executing following to make it permanent.

sudo sh -c 'echo "/usr/local/lib" > /etc/ld.so.conf.d/opencv.conf'
sudo ldconfig