How to identify small part on pipe of a vehicle engine using Machine Vision/Object Detection

TensorFlow General Discussion
,
1

This is image of a vehicle engine on assembly line. The pipes must be fitted with clips. during assembly.

1666334197107 - with marks - 2
1666334197107 - with marks - 21304×878 182 KB

I want to detect if any pipe is missing clip. worker will user tablet / camera and take video feed. during video, if program is able to detect missing clip, it should alert with name of pipe that is missing clip.

My questions to you are

  1. What MV / TFOD algorithms/mechanism will be suitable for this problem statement.
  2. How to train the models?
  3. Should I take sample pipes - with clip and without clip and take several images of them outside engine (on plain background surface) or On engine for training purpose.

I am seeking your advice to

  1. How to establish relationship / identify different pipes based on their relative positions in the engine assembly
  2. Any sample / Github repo? that closely matches problem statement?
3

Hi @Haresh_Gujarathi,

Here is an example to train your own OD Model and you can use the following Readme File for better understanding of training custom OD Model, please let me know if this help you out.
Thanks.

4

Hi
Sorry - let me correct myself. I want to detect ABSENCE of the clips on the pipes
we will get such images and train the mode.
Will there be accuracy
How can system inform me that pipe # so and so is missing clip

6

I suppose, if the goal is to identify all places where the clip should be but is not visible, you can get away with training the model to detect just one class. You should draw bounding boxes only in the places where the clips are missing.

8

No the goal is to identify
pipe (there will be many pipes and each one is unique and has id)
which pipe the clip is missing