# AI security system add-on with yolo8 verification

**URL:** <https://discourse.nodered.org/t/ai-security-system-add-on-with-yolo8-verification/90382>\
**Category:** Share Your Projects\
**Created:** [23 August 2024 22:43 UTC](https://discourse.nodered.org/t/ai-security-system-add-on-with-yolo8-verification/90382 "2024-08-23T22:43:28Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![wb666greene](https://sea2.discourse-cdn.com/flex026/user_avatar/discourse.nodered.org/wb666greene/32/6534_2.png) [@wb666greene](https://discourse.nodered.org/u/wb666greene)\
**Post date:** [23 August 2024 22:43 UTC](https://discourse.nodered.org/t/ai-security-system-add-on-with-yolo8-verification/90382/1 "2024-08-23T22:43:28Z")

</div>

This is a much evolved and simplified version of the project I shared several years ago:  
[https://discourse.nodered.org/t/ai-enhanced-video-security-system-with-node-red-controller-viewer/21622](https://discourse.nodered.org/t/ai-enhanced-video-security-system-with-node-red-controller-viewer/21622)

Only minor changes to the node-red controller interface, the installation is simplified because I dropped support for weak hardware that has little chance of running yolo8 models, and most of the pieces are now installable via pip.

Yolo8 is the key to the very low false positive rate I've achieved, four false positives in about two years of 24/7/365 operation with 26 cameras. Three of the four false positives happened with the yolo4 verification step, 19 cameras run with yolo8 verification on an i7 laptop with GTX1060 GPU, 7 cameras run on an i5 "industrial PC" with NCS2 yolo4 verification.

The last version that supports "weak" hardware (what I've been running) is here, but installation and setup can be pretty difficult:

> **[GitHub - wb666greene/AI-Person-Detector-with-YOLO-Verification](https://github.com/wb666greene/AI-Person-Detector-with-YOLO-Verification)**
>
> Contribute to wb666greene/AI-Person-Detector-with-YOLO-Verification development by creating an account on GitHub.

I'll be making no more changes to this, although I'm still running this version for the foreseeable future

The new version which I will continue to try and improve with some "newer" models (Yolo10, YoloNAS, MobilenetSSD\_v3) can be found here:

> **[GitHub - wb666greene/AI-Person-Detector-with-YOLO-verification-Version-2: The purpose of this project is to add AI "Person...](https://github.com/wb666greene/AI-Person-Detector-with-YOLO-verification-Version-2)**
>
> The purpose of this project is to add AI "Person Detection" to existing security DVRs and or standalone net cameras. The design goal is to be an "appliance" running 24/7/365 and to make the "false positive" rate be as close to zero as possible. False negatives are of little consequence if the frame rate is decent.

Currently it supports MobilenetSSD\_v2 for initial AI detection using either Corel TPU or openvino 2024.3 CPU AI and does yolo8 verification with either openvino iGPU or NVidia cuda capable GPU. The real work is done in python, with node-red providing a simple but effective user interface.

You can see the system in action as a solicitor distributing "flyers" walks from my neighbor's yard to my mailbox and then across my yard to the next house:  
[https://youtu.be/XZYyE\_WsRLI](https://youtu.be/XZYyE_WsRLI)

---

<div class="post-metadata">

**Author:** ![wb666greene](https://sea2.discourse-cdn.com/flex026/user_avatar/discourse.nodered.org/wb666greene/32/6534_2.png) [@wb666greene](https://discourse.nodered.org/u/wb666greene)\
**Post date:** [31 August 2024 18:22 UTC](https://discourse.nodered.org/t/ai-security-system-add-on-with-yolo8-verification/90382/2 "2024-08-31T18:22:09Z")

</div>

I made a really nice version2 update. I discovered that cuda12 is now pip installable like openvino making the installation much simpler!

Ubuntu 22.04 Nvidia driver 535 supports cuda12 so life is good, Should still work on 20.04 too but I haven't test the cuda on 20.04, only OpenVINO since the most cost effective solution is using a Intel processor with iGPU -- I've found "closeout" and "refurbished" laptops that cost less than most Nvidia graphics cards. This has also made me lose interest in "IOT class" systems.

Since all the drivers are available for Windows, it should be possible to run the system on Windows or WSL. You will lose the "housekeeping" functions provided by node-red exec nodes launching shell scripts. This is not something I'm interested in, but raise issues on the github and I'll try to help out. I suspect most issues will be difference is file naming conventions, especially case sensitivity.

At the bottom of the github README is some performance results on a "closeout" i3 laptop for an overnight test with four 4K rtsp cameras.