Hello Carlos and Sam
That book on "CCTV for Wildlife Monitoring" looks excellent. Although it is somewhat off at a tangent of my actual work this looks way too useful not to have. There are cheaper second-hand copies on Abebooks and eBay and also on sale at NHBS, £20.98 incl. UK P&P (https://www.nhbs.com/cctv-for-wildlife-monitoring-book).

An interesting interview with the author, Susan Young: https://blog.nhbs.com/author-interviews/interview/cctv-for-wildlife-monitoring-an-interview-with-susan-young/

Meanwhile, I have come across a research project by the author of that book using technology that looks very close to what Sam is trying to do: "CCTV Enables the Discovery of New Barbastelle Vocalisations and Activity Patterns Near a Roost - We monitored a barbastelle maternity roost for four months using a portable CCTV system, time synchronised with ultrasound recorders."
See more here: https://www.researchgate.net/project/Using-CCTV-to-link-vocal-repertoire-with-behaviour-in-bats

If the above does not do it, I think you should contact the author and get her advice on what would be a suitable setup for you. That's what I meant in my earlier post about maximising use of previously developed solutions. It will just save you a lot of messing about, especially if you are not a technical expert yourself.

I hope this is of some use. Do post an update when you have worked out what equipment to use.

Good luck,
Joachim Neff

Hello Claire, 

Engineer at Wild Me here. We would love to start a conversation about a Wildbook for rhinos. 

Lets talk about citizen science and computer vision for identification possibilities. I'm curious about your current data set and the identification tools you are using as a starting point. I'm happy to talk here, or you can email our team at [email protected].

Hi Claire,

At the BearID Project, we are working on a similar problem for brown bears. We are currently using machine learning methods developed for human facial recognition (like Google FaceNet). We got some ok initial results, but now we are running up against small data issues. The method for human faces were trained with millions of images of hundreds of thousandes on individuals. We have a few thousand images of about a hundred individuals. We plan to investigate other methods in the future.

It will be great to keep in touch to see what methods you will be using.

Ed

Hi David,

Kate Jones suggested trying our their Bat Classifier tool - deep learning tools for bat acoustic signal detection. We've quite a few members here who worked on it if you've got questions (@Rory_Gibb, @snewson, @robin_freeman as a start)

We have a method for finding bat calls that is super quick, free and all code is open source https://t.co/52FPZHLyQU

— Kate Jones (@ProfKateJones) December 21, 2018

Steph

How about using Kaleidoscope and the cluster analysis tool?  Works fine, and is super-quick..

Ta, Carlos

Hello Ollie,

SODA is a recently launched suite. It is in the development phase. We have with us call libraries with multiple (40+) calls for 10- 12 species of birds. A research scholar is on the job for classifying at the species level.

Our interest is equally on separating the different sonic components (as stated in https://www.wildlabs.net/community/thread/666). It is different from the link shared in my previous reply. This is where I see the confluence of objectives.

As a TEAM, I believe, we can address the individual objectives faster.

regards

jaishanker

Hi Ollie, 

Great article, thank you!  I mostly work with responsible AI in other contexts, at Doteveryone.org.uk and the Trust & Technology Initiative at the University of Cambridge, so don't have much to offer here, although I am very interested in the topic.  I appreciate your point that many of the consumer data issues highlighted in the 'popular' responsible AI discourse aren't relevant to conservation (some of us have been gathering 'responsible tech' / 'ethical tech' resources in a shared doc, and there's essentially nothing there for conservation specifically - https://docs.google.com/document/d/1SN6hYeKe3eRK6x9D0Sr7GpCA4nirpyo3u68xG1A6NDs/edit ). However there might be some links with humanitarian data practices, which are touched on by the Responsible Data folks at https://responsibledata.io and https://www.fabriders.net/data-literacy-consortium/ or in this recent article https://asecondmouse.wordpress.com/2019/02/20/instability-forecasting-models-seven-ethical-considerations/

Best, 

Laura

Excellent, that good to hear.

No, I didn't have anyone in particular in mind, it's just helpful to have that info in the back of my mind as I talk to different people/projects. I was also wondering whether to include the invite in our next Digest which will be coming out tomorrow, and based on this response I'll pop it in there :) 

Congratulations! This looks like great tech.

It certainly is. Whatever gets the job done and makes it easier for us to find the animals and prosecute the poachers. Also interested in thermal-imaging hardware and perhaps radar. I'm new to them all.

Hello Sean
I just had another thought: industrial machine vision systems come with all sorts of image inspection tools including such things as "edge detection". I wonder if this could be used in conjunction with a normal (not thermal) high resolution digital camera.

The general idea would be to:
1. take photos whilst on the move through a snare-infested area where a high res camera would be able to pick up a thin wire loop with sufficient resolution to make it visible;
2. let attached computer with vision software analyse images instantly for straight or looped edges that would indicate a snare.

What could help considerably is to take pictures in subdued natural light, maybe at dusk or dawn using a strong flash, possibly not just the built in flash of the camera but an external one. On the assumption that most snares are made from plain "shiny" metal, they should show up fairly well as "bright highlights" against the otherwise fairly dull (in terms of specular reflection) undergrowth. It would get more difficult though if the foilage was covered in water droplets as these would reflect the flash light, too.

I admit this is an idea that just sprung to mind and I have no idea how well this may work in practice. If you want to have a go yourself, set up a typical "snare scenario" and take some high resolution photos as described above from a few metres away. Then use free vision system software to process the images on your computer and see what you can find. Try this one, for example, https://www.teledynedalsa.com/en/products/imaging/vision-software/sherlock/

I have used it some years ago (for an industrial application) and found it very useful on early trials as it allows you to use all features processing images from a local folder rather than a live camera. This would be all you need to test whether there is any merit in this.

Sorry if industrial vision applications are not within your skill set but I thought I mention it here in case someone else wants to investigate it. I am not exactly twiddling thumbs otherwise I would be seriously tempted to have a look at this myself (pun intended:)).

Let me know if you need some more input on this,
Joachim

P.S.: A rough calculation on camera resolution:
- snare wire diameter is, say, 3 mm
- target size in image of the wire is at least 6 pixels (to reliably show diagonal edges)
- 20 Megapixel camera image is about 5000 pixel wide
- then you could cover an area 2.5 m wide per image (5000 / 6 * 3)
> this may be a workable distance to cover per image.