Putting faces into facial recognition

[quest100]

I recently upgraded my PC from one that chokes on any AI to one that has plenty of horsepower for CodeProject.ai. I am now using AI on six of my older cameras that do not reliably trigger. My PC is still humming along with typical trigger times of 250 mS. This is with no graphics card in the system

Now I am adding facial recognition to my front door camera. Seems to work, but I want to know how to best associate facial images to names. Some assumptions/questions:

1. Multiple images at multiple angles of each person works best? Does the identification get better with more images?
2. How to collect the images? I plan to let AI collect several days worth of family and friends going in and out. Night and day. The unknown faces images are put into <New/CameraName> folder by BI. I have been manually moving them into named subfolders. Is there a better way?
3. How to add names and images?
   1. It looks like naming within BI only allows one image per face. True?
   2. In CP.ai it lets you select multiple images per name.
   3. (In BI if you hit the <Faces> button it clears any entries made through CP.ai. WTF??? At least ask before deleting)
4. However, once a face has been identified, no new image appear in the folder unless not recognized. These new images can then be manually moved to the named image subfolders
5. It appears that the only way to update the facial recognition for a name is to delete the name and repeat the name with the new collection of facial images. One cannot simply add new images to an existing name?
6. It would be nice to be able to collect both recognized and unknown facial images and add them back into the named collection of images. If only the unknown images are added to the collection the identification may end up skewed? I kind of envision this as a big diffuse blob instead of a compact blob with a few outliers.
7. I would really like be to periodically go through all facial images seen by the camera and either verify or change the name. This is the way many photo programs work for identifying people.

I’m hoping that there is a simpler and more robust method of associating facial images with names. If anyone knows how please let me know.

 

[VictorS]

Thank you, these are exactly the questions I have. I expected to find answers fairly easily, but everything I found was too basic. Registering faces is practically a black box, which makes it hard to even guess what I should try next. One trick I learned is that ctl-double-click on an alert that was processed by AI will bring up the Status/AI window.

When I had four people registered, AI frequently attached the wrong name to a face in a new alert. When I tried to add more images to a registered name, thinking it might improve the quality, I ended up with the new images registered under their file names instead of associated with the name I had typed in, as you can see in the attached screenshot.

People who post that BI stops saving faces might find that the AI responded "a face was recognized" to every new face because the minimum confidence level is set too low - but that is simply my speculation. I don't feel like I understand how all this works, and using it feels like alpha testing.

 

[quest100]

After entering 6 names, each with 10-30 images, I find CP.ai's facial recognition to be a complete failure. Someone in the database walking up to the front door generates ip to 8 faces for identification. The returned identities are all over the map - young, old, male, female, same and different complexions. Less than a quarter of the time the correct person was returned. Results were even worse for an unknown person. Always identified as multiple different and obviously wrong persons.

I am going to let it generate more facial images and retry when I have several hundred images per person. Until then .

 

[wittaj]

Yep. I wouldn't use/trust it for mission critical.

Most of us have found that facial identification is more gimmicky and novelty than anything else. If you have to put in 5 or 10 or 15 or 35 pictures or more of yourself in the system for it to recognize it is you...then you shouldn't expect much.... My success rate was under 5% so I moved on to other hobbies LOL. YMMV.

It can work in certain situations like a business that requires everyone to stop in front of the camera and the camera is at head height. Outside of that, the percentage of being accurate is probably not going to be super high. You will get a lot of false "confirmations" doing a search.

Someone here posted once how horrible it was inside his house identifying his neighbors and others as him. Another guy his kids and wife were being tagged as him inside the house.

Unless you spend the big bucks that casinos and airports have LOL.

As always YMMV and some here use it and say it is acceptable for their use case.

Heck even in ideal situations like a business with the camera at ideal height and optimal lighting it fails...

What do you think of Dahua's "Uyghur Warnings"?

Good luck. Wish you no ill will. I'm not here to argue with people simply to contribute when and where I can. This is also followed by lending my views on current technology and how it impacts all of us. It's undisputed that splitting of the atom changed the world but once again man decided to...

ipcamtalk.com