1
$\begingroup$

TLDR : Is there an AI available that can recognize employees in a factory and tell when they entered and left pre-defined areas?

I work in a factory where we gather cycle time data from various inputs (computer interfaces, bar code readers and RFID tags). We follow both parts moving along the production line and persons working on those parts because they move from stations to stations during their day and can be more than one at the same place to help one another.

Our goal is to know how many people are working on each part at any given time, but the problem is that it takes time for people to check themselves in when they get to their workstation and they can also forget to do it.

Many employees asked me if I could find a way to track them automatically so they wouldn't need to check in everywhere they go. And then I stumbled upon Microsoft Workplace Safety Demo and Yitu System (this one is scary) but they both seem to be a little overkill for my needs.

After learning about these, here is the ideal AI features that I'd need :

  • Can use video feeds to detect new people in the workspace and prompt someone (via e-mail or text message) for identification.
  • Can use video feeds to recognize persons that are already known and document every time someone has entered or left a pre-defined zone (workstation) in the video feed.
  • Won't force me to have a camera for each workstation.
  • Allows one workstation to span on more than one video feed.

As of now, I found nothing available that does this, but I may have missed lots of things because this is not what I am used to work with. So maybe you know something that could help me with that?

$\endgroup$
  • $\begingroup$ Are you looking for a turnkey solution or algorithms & ideas to write your own solution? $\endgroup$ – Brian O'Donnell Dec 20 '17 at 3:21
  • $\begingroup$ I'd rather not reinvent the wheel if it is already available to me. $\endgroup$ – Frank Malenfant Dec 21 '17 at 19:08
1
$\begingroup$

I am definitely not the most qualified person in order to answer this question, but I might be able to give you a couple of buzz words to go and do some more research on your own. A convolutional neural network is usually used for analyzing anything visual. (you can read about it here: https://en.wikipedia.org/wiki/Convolutional_neural_network)

There are a lot of frameworks that allow you to build recognition classifiers that can be loaded into a real-time video feed and give you feedback when it gets a hit.

Some of the ones that I know of are:

With Caffe being the most popular.

Some articles on how these work:

Hope this is at least able to get you on the right path.

$\endgroup$
  • $\begingroup$ This is very interesting! YOLO v2 seems very impressive. So what I'd need from there would be to (1) detect the person, (2) send the image to a face recognition software and update the information when it returns something (when they face the camera), (3) extrapolate the persons's feet position on the floor, (4) note the date, and person every time they enter/exit each zone. $\endgroup$ – Frank Malenfant Dec 21 '17 at 19:17
  • $\begingroup$ That seems correct for you use case, yes. $\endgroup$ – Joshua Terrill Dec 22 '17 at 22:39
0
$\begingroup$

training a model to identify new persons from video logs seems to be a daunting task.

You will be needing lots of data and computational power for building such a model. and there is relatively low work going on videos due to the amount of computational power required , even training a simple video classifier with reasonable accuracy requires lots of expertise and resources (data and computing).

Having said that recognising persons from videos and "new" person identification seems to be even more difficult , you should definitely read some papers on video classification using convolutional neural networks to get an idea of how hard the problem is. frameworks are only mere tools which give you helper functions to build neural networks , so choice of frameworks is not a question.

The real question is are you sure you want to train a model (CNN) for this task? Can use video feeds to detect new people in the workspace and prompt someone (via e-mail or text message) for identification

for this you should be training a convolutional network on the factory videos dataset , i.e you the cnn trains on frame level for the videos. and recognizing and classifying "known" and "unknown" faces from videos is a research problem in its own right

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.