Saturday 21 December 2019

Birdsy - Monitor wildlife and auto classify bird species

I recently added a wifi IP camera to add to my collection of home-brew wildlife monitoring kit, kindly supplied by Birdsy: .  This IP camera livestreams to the 'cloud' where artificial intelligence (AI) software is used to classify the bird species.  Video clips are saved to your own secure section of the Birdsy website...

Here is a screenshot of some recent captures off my Birdsy webpage:
A selection of captured, species classified clips as presented on the Birdsy webpage
You can that the bird AI is fairly good, it struggles with non-avian species, but given the company is called 'Birdsy' we can let that go...

As with many internet CCTV streaming applications you can review your live camera from anywhere with internet access and a web browser or via a dedicated phone app.  I can now watch my camera live at work too now...

My Birdsy camera setup
The 'value add' is its auto species classification, which is prett neat.  This is the first Goldfinch it classified, I downloaded it direct from the Birdsy website and uploaded to YouTube:

A mouse or two regularly compete with a Robin, the most frequently classified visitor.

As most people will set this up outdoors, you'll need a decent outdoors wifi connection, and access to a power socket.  I've run a 10m DC extension cable to reach the nearest plug socket (amazon link here).  I already have good wifi coverage outside for half a dozen or so other home-brew cameras.

10m 12v extension cable

The Birdsy camera is configured fairly easily via a smartphone/tablet app. You can dial up or down the resolution of the videa stream, presumably to adjust for better/poorer  wifi signals.

It *is* possible to run this with a dedicated wired ethernet connection, this would require either an additional network cable to be run, or once PoE (power over ethernet) setup with the network and power split out (power dropped to 12v) at the camera end.

Here is a screenshot of the configuration app:

IP camera vs home brew
Until recently, all my nature-watching has been with home-made setups using a Raspberry Pi as a processing unit, with either an attached webcam or raspberry pi camera module (or both).  A cursory glance around the web will find many people using IP cameras like this one from Birdsy instead... so what's the difference?

An IP camera, or 'Internet Protocol' camera is an all-in-one device that connects to a wired or wireless network and generates a video stream.  Camera configuration is usually achieved via a smartphone/tablet app or web browser interface.  The camera itself usually does not deal with video storage or motion capture (I'm oversimplifying, as some do bits of both).  They exist to provide a video 'stream' to some other application, e.g. CCTV motion capture software, and/or dedicated storage device, eg a digital video recorder/DVR.  IP cameras often have built in day/night modes, infra-red cut switching (IR cut) and often an integrated microphone.  They also tend to be expensive.

Most of my home made kit is Raspberry pi minicomputer with attached camera module (like the camera bit from your mobile phone).  All of the functionality mentioned above is possible but require fiddly configuration, the addition of extra components, e.g.illumination, IR cut, microphone and some programming (depending on your aims this may be minimal).  While more fiddly and requiring more technical insight, this approach costs a lot less to setup.

The cool thing about the Birdsy camera, isn't the camera, its the cloud-based image recognition system that categorises the bird video clips based on bird species being filmed.  Given that I get a lot of non-avian species I'm hoping that the AI algorithm will be extended to include non-avian species.  I'm sure there are research applications to auto identification of bird species.

As this is the first dedicated IP camera I've had to play with I plan on investigating other creative things than can be done with the video stream, so more at a later date..


  1. Looks like you need to start an AI project using your Python skills:

  2. Maybe, I think it should be fairly easy to make something that distinguishes between 'bird' vs 'not a bird'. The Birdsy algorithm is good as it goes that one step further... I would prefer it to be done locally however but since cloud-based AI is their USP, I doubt that's going to happen ;)
    It is a good motivator to figure out how myself... so many projects, so little time !

  3. Off-on-a-tangent: but take a look at this video. It reports on an interesting AI experiment which ran millions of times showing how the 'players' developed unexpected strategies: