Nov 11, 2022

Help detecting snakes using AI

For those of you with programming experience, I have some questions, if you'd be so kind as to read them.

I want to work on a project where I train a camera to detect snakes in real time and track them while within the camera range. So, pretend I have a backyard here in AZ and I suspect there are snakes, but there are also other critters running around, so if general motion detection is recommended that wouldn't help.

I don't know anything about programming so before I consider investing time or money, I'd like to know if this is feasible/how it works.

Is the the type Of project I can tackle using the online AI programming courses? My thoughts where of using youtube videos of snakes for this project.
1. How difficult would a project like this be for someone with no programming experience. What language would be used?
2. If I paid someone to create the basic program how much would do you think it would cost?

Thank you

Start Discussion

34 Comments

RS

1y

let me know if this is a stupid idea. Thanks

O

1y

You can use AWS to do this. Check out AWS Sagemaker and AWS Kinesis.

RS

1y

I think this is what I'm looking for but will t be feasible to learn to use these programs?

Anonymous Monkey

?

1y

Most Helpful

Just a quick brain dump:

You wouldn't use video, you'd use still images that you pull from that video. So it's not a video problem, it's an image problem. If you want to sound really professional, it's an "image processing" problem
Your biggest challenge will be training data. You'd need all kinds of pictures of snakes in all kinds of environments and positions. You'd basically need to recreate all of the potential ways that snakes could be in real life at any time (day, night, big snakes, small snakes, coiled, extended, on the floor, on a rock, etc). If you think about all the permutations of all those requirements (e.g. on a rock, coiled, at night; on a rock, coiled, during the day, etc.) then you should realize that you need A LOT of training data. Tens-of-thousands of images or more.
Your aim should be to create an algorithm that removes bias and can act independently as your snake spotter. This has been hard to do with the exact kind of "animal spotting" problem you describe. There is a famous example where a researcher did the same thing for a type of endangered wolf. They thought their program worked, but what they really did was accidentally create an algorithm that detects snow because all of the known photos of this wolf (i.e. the training data I'm describing above) contained snow.

I will leave it up to you if this is something that you really want to pursue.

IncomingIBDreject

AM

1y

Your biggest challenge will be training data. You'd need all kinds of pictures of snakes in all kinds of environments and positions. You'd basically need to recreate all of the potential ways that snakes could be in real life at any time (day, night, big snakes, small snakes, coiled, extended, on the floor, on a rock, etc). If you think about all the permutations of all those requirements (e.g. on a rock, coiled, at night; on a rock, coiled, during the day, etc.) then you should realize that you need A LOT of training data. Tens-of-thousands of images or more

Just want to point out that cameras can collect audio data as well, which would be more informative and require less data than purely relying on visual cues.

Array

Anonymous Monkey

?

1y

You should be too busy with therapy to be posting here, son.

IB

1y

how tf are you gonna detect snakes with audio? also audio data is hard to work with and requires a good bit of math if you want to train a model based on audio inputs.

RS

1y

GoingToBeAnMD

Just a quick brain dump:

You wouldn't use video, you'd use still images that you pull from that video. So it's not a video problem, it's an image problem. If you want to sound really professional, it's an "image processing" problem

Your biggest challenge will be training data. You'd need all kinds of pictures of snakes in all kinds of environments and positions. You'd basically need to recreate all of the potential ways that snakes could be in real life at any time (day, night, big snakes, small snakes, coiled, extended, on the floor, on a rock, etc). If you think about all the permutations of all those requirements (e.g. on a rock, coiled, at night; on a rock, coiled, during the day, etc.) then you should realize that you need A LOT of training data. Tens-of-thousands of images or more.

Your aim should be to create an algorithm that removes bias and can act independently as your snake spotter. This has been hard to do with the exact kind of "animal spotting" problem you describe. There is a famous example where a researcher did the same thing for a type of endangered wolf. They thought their program worked, but what they really did was accidentally create an algorithm that detects snow because all of the known photos of this wolf (i.e. the training data I'm describing above) contained snow.

I will leave it up to you if this is something that you really want to pursue.

If I don't have the capital now, I'll have the capital some day... I'm doubling down on this.

I was thinking if it's long and slithering its a snake. Then make sure it doesn't detect a hose because the hose wouldn't be slithering. I'd only be looking at the snake in my lawn. If I used something like infrared, I think I could. I think Sagemaker will work

How easy would it be to build an algo/ how much would it cost?

Anonymous Monkey

?

1y

It all depends on who you hire. I have a team of absolute killers that does this type of work for my function but I pay them quite a bit to be good.

O

1y

Download snek from internetz, train machine learning model on snek, run object detection preprocessing to output snek and objectz, run snek detector on output to detect snek.

Can be done, yes, but is this seriously worth his time?

(Also, I have no idea why I decided to go cacographic on you... Seemed funny at the time.)

Anonymous Monkey

?

1y

User name checks out.

johnny-mnemonic

PE

1y

May be useful to look on inaturalist.org. Millions of pictures of snakes spotted in the wild on there

Authored by: Certified Private Equity Professional - 1st Year AssociateCertified Private Equity Pro

IB

1y

To be fair the best way to detect a snake wouldn't necessarily be through static image processing given that they are very difficult to distinguish from their environment but more like through motion detection.

Anonymous Monkey

?

1y

I appreciate your thoughts but let's not do Level 1 thinking here:

The kid said he wanted video, let's do video
I've been answering with the assumption that "snakes" may be a euphemism for other things he may want to detect or create an algorithm for. That's fair play. I want people to realize that writing algorithms (any algorithm) is fucking hard.
Motion detectors are going to give you all sorts of false-positives and you'd need something very, very high end to detect a snake moving. I used to work in physical security with cameras, access control, stuff like that. The good stuff always costs a ton of money.
If you did this as a business you could have people set up their own cameras and then your algorithm takes over from there, get recurring revenue from the subscription to your algorithm hosted on your own servers. That's a completely defensible model, no one gets your IP. With motion detectors, any setup you make people can copy by buying the same stuff - a horribly indefensible business model. Why would you want that?

I appreciate your thoughts

RS

1y

Correct as noted above big fish are the end game, but I figured snake detection would work too and be something I could try out.

Basically my thought was if I can't capture a slithering snake from a picture. As t moves the snake slithers and its body oscillates side to side. The video angle would be static and Arial. I guess I don't need AI for this since this will always be constant. I just want to detect the moving image on a screen. Looks like Python with OpenCV or tensorflow might work.

PT

1y

Nah that’s bs. Use python skilearn and pandas with keras. Everything already done. At that point you just need to

1) find a database with the training data

2) upload the database with your images

BigKahunaBanker🏄🍹🍔

Senior Neanderthal

IB

1y

Bro, studied, educated SWEs routinely fuck up tasks like this even with million-dollar budgets. What makes you think you can do this by yourself with no education, no money, and no incentive beyond "but I wanna know"?

...and the Truth shall set you free

Pug

O

1y

Every SWE was once someone who didn't know how to code

“The three most harmful addictions are heroin, carbohydrates, and a monthly salary.” - Nassim Taleb

IB

1y

You can probably train a support vector machine to analyze stills from the video capture and identify a pre-determined list of critters.

Array

IB

1y

Why do you wanna spot snakes lol

?

1y

This reminds me of the hot dog/not a hot dog episode of Silicon Valley with Jin Yang

CB

1y

Architecto dignissimos quia esse. Omnis assumenda perferendis est omnis modi neque.

Maxime quas explicabo sed est voluptas architecto. In illum adipisci commodi ad. Velit vitae cumque dolor tempore vel debitis. Perspiciatis et voluptates blanditiis sunt.

Officiis velit voluptatem odio harum. Quia nobis repudiandae delectus.

Laboriosam ut eum ducimus. Vitae temporibus voluptate sapiente vero odit quo neque illo.

Borrow at 3, Lend at 6, Golf at 3

Sir Desmond Glazebrook

IB

1y

Reiciendis illo tempora tenetur. Consequatur mollitia impedit ut delectus placeat ipsa quia sit. Fugit excepturi fugiat dolorem natus.

Quod fuga dolorem dolores ut nemo omnis est. Provident laboriosam omnis odit voluptatem quia doloribus dolorem earum. Atque corrupti quia ullam doloremque praesentium. Quia ipsa est voluptatem. Quas aut et omnis sed aut quasi dolorem et.

Start Discussion

Career Resources