BigJinge's profile - activity

Thanks muchly.

I'm using Python.

The Integrated Catflap : Fine Tuning Cascade Hi, I've built into a stock catflap, a prey sensing Raspberry Pi OpenCV ca

Thanks, that's the great help. I have been doing some reading about deep learning since posting the original question a

I'm looking at neural nets now, so I'll add that to the list. :)

Thanks sjhalayka. I started it a couple of years ago but due to life commitments have only spent 6 months or so on it t

In answer to the person who made the quote about the cascade being a binary classifier. The no cat, cat, cat with prey i

The Integrated Catflap : Fine Tuning Cascade Hi, I've built into a stock catflap, a prey sensing Raspberry Pi OpenCV ca

Dear Hive Mind,

I've been been doing test haar cascades and have found the right values to give the results I need.

The problem is with syntax of detectMultiScale in Python for OpenCV 3.1. I need to put a minimum and maximum value for size and I'm not sure which order to put them in / flags to use.

My values are :

Scale factor = 1.05 Minimum neighbours = 3 Minimum size 48x48 Maximum size 120x120

I've seen an example that looks something like this :-

detectMultiScale(img, faces, 1.1, 3, CV_HAAR_FIND_BIGGEST_OBJECT, Size(30, 30), Size(200,200)) and regarding CV_HAAR_FIND_BIGGEST_OBJECT it says you may or may not require it.

So, can you throw some light on what is the correct detectMultiScale syntax I should use.

Many thanks

BigJinge

In theory the benefit of only having mouse parts detected rather than the cat per se is there may be reuse of the cascade XML for people who don't have the same cat facial features.

I got hold of Cascade Trainer GUI by Amin Ahmadi. I'm going to use that to make both a Haar and LBP cascade to see which gives better results.

Of course, if we can help each other, that would be great.

Best wishes

BigJinge

Hi Nicoenz,

Sorry for not getting back to you earlier but I wasn't notified that someone had replied to my question.

If you are getting results then I take it you've built the cascade. I'm at the point where I now have nearly 10000 negative pics and am going to build mine.

I had read that it's best not to include blurry pics in the positives as the recognition isn't great. I was going to go for the "prey in the mouth" front under shot but as I found over the months of the picture capture being active that the cat very rarely approaches straight towards the flap. I've got plenty of side head tilts, face down etc. What has been noticeable within the prey pics is mouse feet and the shape of the mouse at different angles. I'm going to highlight those.

Hello,

I've built into a standard Petsafe cat flap, the innards to be able to take a photo underneath the cats mouth, to see if has an item of prey in it.

If it does then the flap will trigger a servo and lock the flap.

I've had the flap recording since September each time the cat goes back and forth through it. I now have 500+ pictures of prey in different positions in the cat mouth.

The plan is to create an Open CV Haar Cascade from these images which Open CV will use to lock the flap.

Question is, is using a Haar Cascade the best option for recognition? Would LBP be better?

Many thanks

BigJinge