{"id":1204,"date":"2026-09-02T17:13:55","date_gmt":"2026-09-03T01:13:55","guid":{"rendered":"https:\/\/salemdata.net\/johnpress\/?p=1204"},"modified":"2026-09-08T13:14:14","modified_gmt":"2026-09-08T21:14:14","slug":"build-your-own-artificial-intelligence-camera-system","status":"publish","type":"post","link":"https:\/\/salemdata.net\/johnpress\/?p=1204","title":{"rendered":"Build Your Own Artificial Intelligence Camera System"},"content":{"rendered":"
<p>I have added artificial intelligence (“AI”) to my home surveillance camera system.<\/p>\n<p>No cloud service.<\/p>\n<p>No fees.<\/p>\n<p>No meters.<\/p>\n<p>Everything stays on premises and is totally under my control.<\/p>\n<p>The system I have designed is quite simple:<\/p>\n<ul>\n<li>you need a Raspberry Pi 4B (<a href="https:\/\/www.adafruit.com\/product\/4296">Adafruit<\/a> \\$120 – \\$190) or better,<\/li>\n<li>a Google Coral USB accelerator (~ \\$120, used to be \\$60),<\/li>\n<li>surveillance cameras that expose their rtsp feeds, and<\/li>\n<li>[optional ]Moonfire-nvr – a Rust-based video recording system developed by Scott Lamb, a former Google engineer.<\/li>\n<\/ul>\n<p>I’ve been using <a href="https:\/\/github.com\/scottlamb\/moonfire-nvr">Moonfire-nvr<\/a> since 2018 and have had up to 13 cameras surveilling my two adjacent properties.\u00a0 \u00a0Moonfire has been rock-solid.\u00a0 I cannot commend\u00a0 Moonfire enough.\u00a0 The reaction I get from people when I suggest Moonfire is that they want something that has AI detection.\u00a0 So I set out to see what I could assemble as a companion AI paradigm, keeping it simple and serviceable.\u00a0 I was thinking I might have to integrate into Moonfire, but I have decided to not pollute Scott’s code and live with some easy to configure and deploy scripts.\u00a0 Here’s what I have come up with.\u00a0 Yes, there are a lot of parts… that’s the beauty of owning the entire tech stack.\u00a0 And, everything I have added here are scripts, so you’re off to the races.<\/p>\n<h2>Background<\/h2>\n<p>The detection model I use with Google Coral expects an image 300 \u00d7 300 pixels in size, which is converted into the tensor the model processes. So I started from there and worked backwards because I want to minimize distortion where some software\u00a0 scrunches down and distorts the region in your video to fit the 300 x 300 format.\u00a0 You may not see how much distortion occurs.\u00a0 That’s the beauty of proprietary software; it hides the ugly shortcuts it might take.<\/p>\n<p>I run most of my cameras at their highest resolution — I want to have a chance of reading a license plate, or capturing the scar on someone’s face, or getting the pattern on their socks that readily identifies them.<\/p>\n<p>Yes, one thief served 6 months for stealing my contractor’s concrete saw.\u00a0 See https:\/\/salemdata.us\/videos\/sawthief.mp4<\/p>\n<p><img class="alignleft size-full wp-image-1216" src="https:\/\/salemdata.net\/johnpress\/wp-content\/uploads\/2026\/09\/20260902_175454_Wed.png" alt="" width="614" height="384" \/><\/p>\n<p>I capture a full-framed image from a video camera using ffmpeg, a set of software tools for video, audio, and other multimedia files and streams.<\/p>\n<pre>export PASSWORD=[*FILL IN*]\r\nffmpeg -rtsp_transport tcp \\\r\n-i "rtsp:\/\/coral:${PASSWORD}@192.168.1.132:554\/h264Preview_01_main" \\\r\n-ss 2 \\\r\n-hide_banner -loglevel error \\\r\n-frames:v 1 \\\r\n`date +'%Y%m%d_%a_%H%M%S'`_Court180_full.png\r\n<\/pre>\n<p>This produced an image 4608\u00d71728 which is the native resolution.\u00a0 I then load the image into a simple editor written in Python, <strong>roi_select_snapshot_resume.py<\/strong>, I created where I\u00a0 can place 300 x 300 pixel squares, or multiples thereof, e.g. 600 x 600 or 900 x 900 and then define regions of interest.\u00a0 It’s like using a cookie cutter on a rolled-out dough,<\/p>\n<p>Here’s a video (no audio) showing a sample session creating ROIs:<\/p>\n