I am working on a project. In this project I need to use a robotic gripper to gripp a flange. This flange will be detected and located in a monocamera. This flange contour comes from a FOUP handling flange. What I need to do is to detect the most outer edge. Because that edge is determined by a CAD model. The internal edge is not useful.
I have already tried to detect the edge by using bilateral filter and Canny detector. But this works not very well. Because the internal edge is more clear than the outer edge. The result is, there are many edges at the contour. And that contour is also not continuous.
The following 3 images are
- original image
- outer edge, I need to detect
- edge image by using Canny detector
Can anyone give me an advise, how to detect the outer line accurately?
I would try thresholding the image (by trial and error) and then apply the circular hough transform to see if proper detection of the rings can be achieved. Based on the results you get maybe I could suggest other refinements
Regards
Thank you for your reply.
What you advised, is exactly I have tried. And that works very well. Because I need to use robot arm to grip that flange. The camera needs to be calibrated. So I can calculate how many pixels for how many mm.
I have already used the ring information to segment this flange into 8 pieces. I use the circle middle point as a start point, circle radius as the width and height.
You may see, there are 8 notches. Each segment has a notch. Notch is where gripper to grip.
Now it comes hard. Because there is also a inner edge of this flange. The outer edge is not easy to detect. Also, the right side of this flange, the outer edge is not very obvious.
Regards
I'm sorry. I don't unterstand you quite clearly @ Zewen Liu
What do you mean by the other simple musks? Which filter is worth to considerate?
First you would try histogram equalization. After that, you can use closing and opening operators to impress the edges. And then, you can try edge detection methods like Roberts, Sobel, Prewitt, Kirsch, Robinson, Marr-Hildreth, LoG and Canny. Finally, You can use region growing algorithm using special seed point (out of the wanted region) to detect outer line of the field.
Regards.
Jialiang Yin
If you can use the calibration from the camera and this information can be used to solve your problem, then instead of trying to detect the outer edge, try to detect with as much precision the inner edge. From there, you can calculate the offset to the outer edge.
Try:
1) Do binary erosion/dilation to clear the noise ( 2.png is quite noisy)
2) Do an additional elimination (if necessary) of pixel clusters on the outer segments of the image that do not fit a threshold x (say if less than 20 pixels on either direction or diagonals eliminate)
steps 1 and 2 should clear up 2.png further so as to leave long linear segments along the borders. Based on 2.png, I would concentrate on the inner line since given the lighting conditions it is the best candidate for exact determination and then use the pixel offset and the physical dimensions.
For determination of the inner edge
3)Do a horizontal scan and calculate the distance from the inner circle to the first instance of the line detected (using 2.png and the measurements obtained from your inner circle calculation). store the frequency of the offsets
4) calculate the angle of each line segment detected.
5) for the horizontal line the lowest value of distance from the center that has the highest frequency can be taken as the inner edge. Calculate the offset to the outer edge physically and compare with the result
Now repeat with the diagonals and repeat in a vertical scan to get the top and bottom parts
Let me know if this works or give me the results to see if there are other alternatives.
Regards
simply apply a Horizontal and vertical line kernel to get the gradients and then calculated the overall gradient: sqrt(gx**2+gy**2). The kernel for horizontal line like kx=[-1,1]
Something that I proved and worked well is apliying a gaussian filter and sobel x and y, it delimites well edges, or a canny algoritm that uses derivates but the cost effience could be important. The filter is inportant because will help you to separate low frequencies from high frequencies. Also Try analizing the image in gray scale, it will give you an aproach to decide wich algorithm use.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing