OK, hang with me on this but I was wondering if there was any way to get access to the Face ID SW for use in other systems? I want to see if I can get Apple’s Face ID to recognize individual sheep faces. Reason is I want to be able to take a picture of a sheep with my phone and have it then pass the ID info to LambTracker for lookup of the details about that animal without having to be close enough to read an ear tag or scan the EID tag. I know most of the sheep by sight but I was thinking that would be a really cool addition to LambTracker esp. for shepherds with larger flocks that don’t know all their animals.
Future idea is to combine with AR glasses (wish the google ones had made it into service) and some other wearable technology for hands free or chorded buttons (since so far voice input via Siri still sucks) to move between info screens about individual sheep.
I know I can add a second face to the system. But I am not sure I can get the info out of FaceID for my own use.
I’m already investigating OpenCV and some other stuff written in Python but I thought I could do a quick prototype using FaceID faster than getting the other stuff running.
I looked into it but it only returns an authenticated or not as a boolean. I need to return which face did the authentication. Worse, FaceID only allows 2 faces to be registered. Back to the drawing board (or in this case better google searches now that I’ve had some coffee)
If you were wanting to go the Apple route, I think it would be through CoreML and training the phone to recognise sheep. The learning curve would be nasty though.
FaceID is going to be pretty locked down for security reasons I think.
Yeah the more I’ve looked into it the more I think you’re correct. I think this has to go on the someday/maybe list for now. Too much else to work on that is more critical.
I think the idea with FaceID (and TouchID) is that you cannot get any data out of them other than the boolean yes/no identity confirmed/not confirmed. If I recall correctly, that’s to keep any biometric information from being exhilarated from the system.
Instead of Face ID I would look into Photos facial recognition. If not there, possibly some other software for image analysis. Next question is how far away does facial recognition work? If it doesn’t recognize a human face at the distance you need, there’s not much hope recognizing a sheep’s face at that distance.
Do you identify your sheep by face or other attributes? We were at a fiber festival and sheep dog trials earlier this month and I would have a hard time telling some of the sheep apart.
BTW Photos can get confused…just last night I marked as misidentified a few hundred photos of our younger kid (person not goat) who it thought was the same as the older.
No clue but very good point. During lambing I can paint numbers on the side of the sheep and then it’s a simple lookup if it recognizes the number but they wear off within a week or so. If I coat the sheep
Another very good point. When I try to analyze how I know which sheep are whom I can tell differences in their faces. I can usually get to within 10 feet of even the most skittish sheep so I am a lot closer than across the pasture. I can’t tell some twins apart unless I read the tags though. I can also tell individuals apart by their gait, how they walk. I have 2 adult rams that are similar in size and behavior and I can tell them apart from 300 feet away because one swaggers and one marches. I can’t tell by looking at faces at that distance because their horn shape is similar. I can also tell some individuals by their baa. Not just that they are the ones that answer me when I baa at them but that they sound different. But baas change with stress and age. A lamb baa with a certain tone in it can pull me out of a dead sleep in the summer when I sleep with my window open. It will have me out of bed and pulling on my pants before I even really know what I heard. I have no clue what I sense in that situation but both of us have done that and been rolling downstairs before we even were really awake. A few I can tell apart because of who their buddies are. I know that there is a gang of 3 teenage rams that all hang out together. I can tell 2 of them easily and I can be 99% certain that the third is the other gang member when I see 3 rams of that age together. Usually getting into mischief but that is another issue.
So my identification of sheep is multipoint with many data sources and probably a lot that I am not even aware of.
First article talks about Cambridge and determining stress in sheep. That is also where they studied sheep and discovered that they can recognize roughly 150 faces as individuals whether human or other sheep. One reason sheep learn their shepherds.
Second is exactly what I am thinking of but my use case is more interactive. THANKS for that link. Most of the cool sheep research from NZ is not allowed out of the country (I’m looking at you CARLA testing) :sigh: but that one might be because so few folks will use it.
And of course we haven’t even gotten into the issue that my sheep are black and if machine learning systems are trained on white sheep they might not recognize black ones as sheep.
And again, MPU comes though with more info and potential ideas than I could come up with on my own.
Quick Q: are you already using some sort of sub-dermal ID tagging chip system? We use Trovan microchip ID systems (ID-100US) for wireless ID in our populations, but you have to have a hand-held scanner device… which may not be what you want since you’re looking to implement some sort of AR I put down the line.
No, sub-dermal is not allowed in food animals except in the ear cartilage and that migrates. I am using ear tag EID but with our own handheld scanner my husband designed.