How to replace the CNN of the wrist camera with one of my own design

ruihao_xie

I have a UR3 robotic arm and a wrist camera, which can realize object recognition on the teach pendant, but I hope that the recognition algorithm can be changed.

I designed a CNN algorithm and hope to use it.

bcastets

You can retreive the image of the last screw shoot taken by the camera and process it.

Check this post:
https://dof.robotiq.com/discussion/275/wrist-camera-live-image

bcastets

@ruihao_xie
You can only get the imagine from the camera. Once you have this imagine you can analyze it on a PC using the technology of your choice. It could be neural network.

There is no CNN involve with the camera.

https://scikit-image.org/ is  nice image processing library.
You also have Opencv. The website https://learnopencv.com/ is good to learn opencv.

ruihao_xie

bcastets said:

You can retreive the image of the last screw shoot taken by the camera and process it.

Check this post:
https://dof.robotiq.com/discussion/275/wrist-camera-live-image

I did it this way. My browser shows "status:  success",and I can see the picture. Thank you very much.

and，second question，I want to change the model of the neural network.
I do not know what to do.

ruihao_xie

bcastets said:

@ruihao_xie
You can only get the imagine from the camera. Once you have this imagine you can analyze it on a PC using the technology of your choice. It could be neural network.

There is no CNN involve with the camera.

https://scikit-image.org/ is  nice image processing library.
You also have Opencv. The website https://learnopencv.com/ is good to learn opencv.

  Is there no CNN involve with the camera ?  How does camera recognize objects ？

bcastets

@ruihao_xie
We have a software analyzing images on the UR controller. It use traditional vision algorithms. We don't use CNN.

ruihao_xie

bcastets said:

@ruihao_xie
We have a software analyzing images on the UR controller. It use traditional vision algorithms. We don't use CNN.

So, can I only use the CNN algorithm externally?

Is there any other way besides this?

bcastets

@ruihao_xie
As you have the image of the camera, you can do what you want with a custom vision algorithm. Use CNN or other technology to analyse the image.

ruihao_xie

bcastets said:

@ruihao_xie
As you have the image of the camera, you can do what you want with a custom vision algorithm. Use CNN or other technology to analyse the image.

I want to use the camera to take pictures of apples ,and I will use CNN to identify it.
and I want let the UR3 arm grab it. What should I learn?
Thank you very much！

bcastets

@ruihao_xie
You have to learn how to analyze image with vision library like openCV. It is quite complex.

ruihao_xie

bcastets said:

@ruihao_xie
You have to learn how to analyze image with vision library like openCV. It is quite complex.

I want to recognize Apple.

If the camera recognizes an apple, how to drive the robotic arm to automatically grab the apple?

bcastets

@ruihao_xie
It is quite complex operation. You may find information about such project on the web. It is difficult for me to explain this in details.

ruihao_xie

bcastets said:

@ruihao_xie
It is quite complex operation. You may find information about such project on the web. It is difficult for me to explain this in details.

Very grateful for your help