100% Local AI Surveillance System - Moondream 1.6B / Python / Mistral 7B

100% Local AI Surveillance System - Moondream 1.6B / Python / Mistral 7B
👊 Become a member and get access to GitHub:
/ allaboutai
📧 Join the newsletter:
www.allabtai.com/newsletter/
🌐 My website:
www.allabtai.com
In this video i create a project where I use the moondream tiny vision model and a live stream to detect different objects / persons and send a e-mail alert with the video clip attatched. Python Automation
GH Moondream:
github.com/vikhyat/moondream
00:00 Local AI Surveillance System Intro
00:44 Flowchart
01:48 Python Code
05:24 How to Setup IP Webcam
05:51 FIRE Test
08:10 BOILING WATER Test
09:49 Conclusion

[@electromigue]

bro, start teaching this kind of stuff to young people in your area and you will become the town hero. I'm not a coder and with Python it is still possible to understand what's happening in the program. Love your vids & your creativity.

[@AllAboutAI]

thnx a lot =)

[@dawidw.6016]

He is teaching already

[@scriptoriumscribe]

wonderful job! i'm going to copy this and tweak it to hunt for ufos!

[@AllAboutAI]

haha cool :P

[@FBHearty]

You're definitively my favorite channel !

[@AllAboutAI]

thnx a lot :D

[@nateb.1276]

super cool video. this is basically what the startup Groundlight is doing.

[@Ricking0]

minimalist wonderpiece, thanks for making this video

[@HyperUpscale]

Great video! 🔥

[@AllAboutAI]

thnx for tuning in =)

[@wildfotoz]

Love the mountains in the background. Looks like you live in a beautiful place!

[@AllAboutAI]

yeah its cold, but nice =)

[@dankoyy42]

you are amazing bro

[@nevildev]

Boiling water test was awesome!

[@AllAboutAI]

thnx :D

[@nerual_dreming]

Bro, it's so amazing engineer skills, u'r NEURAL VIKING!

[@AllAboutAI]

thnx a lot, appericiate it =)

[@enesgul2970]

Harika fikirlerle videolar üretiyorsun.

[@RoombaMaster2020]

Thanks!

[@MilMike]

person of interest vibes ;)

[@bovjer]

Great videos as always mate! I'm subscribing and liking as much as I can. Btw what kind of camera are you using for surveillance? Do you think something like the ESP32 cams or something similar like the ones running on a raspberry pi can be used for this? I do have a beefy GPU, but my budget for camera equipment is in the toilet. Just wondering. Keep up the great work buddy!

[@ereheryeht]

Brother you got some skeeeeelzzz

[@craigrichards5472]

Love the surveillance app, how would you go about detecting if it was not a family member? I think that would be a terrific extension. Amazing mate

[@jackford3069]

You'd probably have to have a few images of yourself and your family members and train the model to recognise them. Training also may not even be required here, not too sure as I haven't played around with it myself.

[@test12382]

Alright I can't take it anymore - ILL JOIN THE MEMBERSHIP

[@AllAboutAI]

welcome :D

[@Tsardoz]

Moondream reduces images to 440x440 pixel squares so no point in high res images. It will letterbox them too. If you want to see small objects you might want to break up the images into overlapping tiles.

[@IamAshishChauhan]

Would love your vlogs if you are going to do it in future. ☺️🙂

[@AllAboutAI]

i might do that :P

[@IamAshishChauhan]

@@AllAboutAI looking forward :)

[@AlvinEvangelista]

I've been watching your content for a while, very good and very useful information. You covered something I wanted to know for a long time. Just wondering if open source has software that allows for vehicle license plate recognition and face recognition as well.

[@AllAboutAI]

thnx mate =) if not to sure about that, havent tested

[@Ricking0]

Yes, check out Home Assistant and Frigate, I believe that is one way

[@widgetman]

Insane! Immediate subscribe

[@user-ei3fw5ct3s]

good stuff !

[@AllAboutAI]

Tnx:)

[@wurstelei1356]

Isn't LM Studio sadly closed source? Anyway good video, thumbs up.

[@hilmiterzi3847]

Be honest, was the hardest part of this video the UI introduction part?

[@Professorfungi]

🎯 Key Takeaways for quick navigation: 00:00 🌐 *Introduction to Local AI Surveillance System* - System demonstration of capturing video and sending email attachments. - Brief overview of the system components and flowchart. - Explanation of the IP webcam setup for live streaming. 01:49 🤖 *Mistral 7B and MoonDream 1 Ty Functions* - Introduction to Mistral 7B function for security logs. - Overview of MoonDream 1 Ty Vision model function. - Configuration of video capture settings. 03:39 📧 *Email Function and Main System Loop* - Explanation of the email sending function using the Mailgun API. - Main loop of the system for continuous monitoring. - Conditional statements based on person detection and email sending. 05:16 🔥 *Testing for Fire Detection* - Switching the system's prompt to detect fire. - Live test for identifying a fire in the video stream. - Evaluation of the email confirmation for successful detection. 07:56 💦 *Testing for Boiling Water Detection* - Setting up the system to detect boiling water. - Live test for identifying boiling water in the video stream. - Verification of the security log and email confirmation. 09:54 🎉 *Project Conclusion and Invitation* - Recap of the project, its ease of setup, and successful tests. - Invitation to access the project via a link in the description. - Closing remarks and gratitude for viewers tuning in. Made with HARPA AI

[@lokeshart3340]

😅 sir i have a request can u make a video on it like I want to create a virtual assistant with vision so I want to use Llava multimodal . When I start the app then it starts storing frames from webcam and a certain interval then I speak to assistant and when I stop speaking the frames will be stored in a folder and last 60 frames will be stored then after detecting that I am not talking it will take all these 60 frames and stack upon each other and make a one image for convenient then this one image containing all last 60 frames in one image will be sent to the modal along with STT as a prompt and then we will get our response from the modal and speak it using TTS. So I am making a vision virtual assistant that uses webcam ☺️😀 this is my idea can u make it? Pls ❤️☺️🙏❤️

[@lokeshart3340]

Like gemini demo real life

[@lbf5984]

Hoping this can be used with zoneminder as well?

[@FamilyManMoving]

Zoneminder captures in motion JPEG - basically a series of JPG files that are replayed like a flip-file card deck to create video. So in theory it will work with moondream if you can feed the images to it in order. You'll probably need to code for MJPEG. I gave up zoneminder because it's sooo far behind others, and it's management and performance over time get worse. I had two sites running it with between 2-8 cameras each. It did a good enough job to put some burglars in jail (two events) and held up in court - or at least the perps plead out after seeing the video (which I converted to H.264). But it's not H.264, which has far surpassed anything else. Most cameras do not output HD-quality MJPEG, so using ZM requires either lower resolution images, or a heavy transcode/rip on the server as the HD feed comes in (CPU hog at scale). So it rips the video into thousands of photo frames. Each of those is an inode on the disk, and you'll find yourself tuning them (and the OS) to get it all to hold together. The chain of daemons to capture, rip, "eventify" and store will add up cycles. It is not an efficient design or system. I liked it and ran it for a few years, but I never stopped looking for something else. Eventually I found and tested a few. Take a look at FreeNAS; that is my current selection. It's pretty cheap compared to your labor. But as always, YMMV. Enjoy.

[@ForTheEraOfLove]

Who knew the Orwellian future would be initiated by our own hands

[@Melvin420x12]

Interesting take, perhaps it's our only weapon against an actual Orwellian future. If the power is decentralized and into the hands of the people, no centralized organization is viable enough to grow big enough in surveillance.

[@CustomDabber360]

sucky sucky long time

[@micbab-vg2mu]

nice:)

[@AllAboutAI]

thnx mate =)

[@akkitty22]

Minimum requirements to run moondream is 6GB VRAM, correct? Sadly can't be run on low end machines yet.

[@AllAboutAI]

that might be correct, havent looked it up

[@fullcrum2089]

You can run using only CPU using llamacpp, but will be very slow and you need RAM that could fit the model.

[@jekkleegrace]

👓🔭

[@rocket-mx6bh]

is mistral even necessary in this structure?

[@John-ki9fu]

reolink already does all this cool tho

[@Ricking0]

sure but mostly person or dog AI detection, this is very customizable and you control the logic, while with reolink god knows where the logic takes place, I can think of 1000 use cases for this.

[@hmoodhsjss7075]

the URL discord invitation doesn't work

[@AllAboutAI]

hmm, ill check asap

[@hmoodhsjss7075]

@@AllAboutAI ty, i just tried it and it still doesnt work, the invitation is expired or i dont have permission to join

[@MrJoaopaulofurtado]

How much to host this in a vps?

[@AllAboutAI]

hmm, not sure

[@yoagcur]

That literally wasn't a cool video. Glad there was no twist at the end with the phone falling into the pot

[@pastuh]

TLDR: Easy, but you should become a paid member.

[@Mephmt]

I am a paid member yet do not have access to the github. Suggestions?

[@AllAboutAI]

just send me a mail or join discord =)

[@user-rn7cp8mr6p]

So you have to turn pc on all the time to make it work? That's BS