About: beaks.live – the software

This is the bird box that is shown at beaks.live. It is on the side of a house in Cambourne, about 8 miles west of Cambridge, in the UK.

Right from the start, the plan was to get it working roughly and quickly and then improve it until it was the best I could do with the crap hardware – this being a £11 webcam connected via USB to a Raspberry Pi 4, which also drives transistors to work the cheapest infra-red LEDs I could find.

Having messed around with RTMP (no one uses it any more) and HLS (I’ll be fucked if I can get it to work) for streaming, I eventually ended up with this system:

The Raspberry Pi takes care of the camera and lighting, uploading the video to the server (a VDS hosted with Mythic Beasts), which does all the heavy lifting of looking for motion and streaming live footage to the many dozens of viewers who are eager to catch a glimpse of beak.

Did I mention the camera is crap? The automatic exposure sets itself to some random level and occasionally flashes up and down twice a second, apparently to relieve the boredom. So the R-Pi has to sort out the exposure, and luckily, you can set most of the camera settings manually via USB. Every 10 minutes the Pi records 5 seconds of video, takes 5 frames and averages the light level on each of them. It then sets the exposure, gamma, and LED levels* depending on whether it needs to be lighter or darker. Or it just leaves things as they are if it’s all hunky dory.

* the LEDs are so dim I just leave them all on all the time now.

It records 5 minutes of video at a time, using FFmpeg (with some video tweaking and normalisation to make the crap camera’s video a bit nicer), which is then uploaded to the server. Funny story – I originally set up the Pi’s exposure setting software so it calculated the camera’s exposure settings from this video – this video which has been normalised. So whatever is coming out of the camera, FFmpeg “fixes” it, and then exposure setting software thinks everything is hunky dory, despite the exposure being so wrong the video is just noise. This is why it records 5 seconds of unfixed video separately to check the exposure. A couple of months later I had forgotten this, and had the brilliant idea of using samples from the 5 minute feed rather than doing a separate 5 second one. I thought the camera had died, until I remembered the normalisation and why I didn’t do it like that originally. I look forward to doing the same thing again in July, September, November, etc.

Incidentally, all this software is written in a mixture of Python and Bash scripting because I am a masochistic lunatic. I love Bash – it’s just mad, with random shit like functions looking like “function my_function () { …” where the ()’s do nothing because you can’t put anything inside them – they are purely decorative.

But I digress. The server has the latest video uploaded to it. It keeps the last 4 uploads so there is 20 minutes of buffer. It deletes the oldest one once it has been processed for motion detection. There is a watchdog timer on the server and the Pi will only upload a video if it’s been updated recently enough. This is to stop the server being filled up with files if it reboots and the processing stops or something. Each 5 minutes is about 100MB.

The motion detection is done with DVR-Scan and hits are processed to generate thumbnails and a static web page. Anything less than 30 seconds long is discarded to get rid of most of the dross. Videos older than 25 hours are deleted so there’s a rolling list of videos.

The live page is also static and uses video.js for the player. The current 5 minute chunk location is obtained using an XMLHttpRequest, then the video loaded with JS. When it gets to the end, the JS gets the next section and plays it with a minor blip for the viewer.

The “live” video is actually always 10-15 minutes in the past because it takes 5 minutes to record a chunk before it’s uploaded and then the server tells the player to play the previously uploaded one so you don’t start watching one that’s still uploading.

It’s a bit like the HLS streaming system, except there’s hideous latency and mine works. If you want to mess it up, right click and choose “show all controls” and then slide the slider to the end. I’ve no idea why I’ve told you that.

About: beaks.live – the hardware

This is the bird box that is shown at beaks.live. It is on the side of a house in Cambourne, about 8 miles west of Cambridge, in the UK.

When I put a camera in this bird box last year, I was not optimistic. Expecting to capture nothing more than the inside of an empty box, there didn’t seem much point in spending any significant sum of money on a camera. I decided to see how well I could get it working for how little money.

Two cameras for £21.66 doesn’t scream quality, but they are able to manually focus down to a few cm. Being cheap and nasty also means they won’t have an infra-red filter on the lens, which means I can illuminate the box at night with a light the beaks can’t see.

I picked one and sawed off the mounting at the bottom, knocked up a 3d printed housing to fit it in the apex of the bird box roof, and fitted some cheap Ebay IR LEDs.

A mess of wires being put into the 3d printed camera mount.
Cheap and nasty does it every time

This is the camera and LED housing mounted in the bird box:

Looking up into the box with the mounting fitted.
Looking upwards into the roof of the box

On the outside is the 3d printed box which holds the interface to the cable that goes into the house and the drivers for the LEDs. I actually had a proper PCB made with a D/A for the microphone but I never wired it up because I’m lazy. That’s why there is no sound. Sorry.

The interface box with unused D/A.

The LED controls and USB for the camera share a length of CAT-5 cable into the house, where they plug into the Raspberry Pi, which has an ethernet connection to the router.

And that’s the hardware. Total cost probably around £75, including custom made PCBs, which were ridiculously cheap. I mean like stupidly cheap – around £5 for 5 PCBs, including delivery from China. Anything clever is done in software, including stuff to improve the performance of the (frankly substandard) parts I used. Next year I’ll replace it with decent kit, including a camera that isn’t shit.

Coming up next… The software