We turned Denise’s camera traps into near-sighted minions!
Denise studies butterfly behavior here in Panama. She came to the GOSH 2022 Global Gathering because she wanted help with a challenge she had studying her butterflies with camera traps.
The PIR sensors picked up the butterfly movement fine (to my surprise!), but she needed to get up close imagery of them, and the camera traps normally have a set focal length for much further away. We were thinking about hacking the lenses, but during the camera trap hacking session at #gosh2022 (there’s no tag for this yet) with @hpy@Albercook , we came up with an even easier fix!
I borrowed @Albercook 's glasses since he was far sighted and connected them to the front of her camera trap, and voila, the camera could see her close up butterflies in great detail!
This week Denise bought some cheap reading glasses (+3.00), and we velcroed them on!
I meant to respond earlier, but this is amazing and beautiful!!! I love the simple elegance of this hack by @hikinghack and @denisedd. I hope you publish this as open source hardware in a paper!
@denisedd: I am curious, do you only use the camera traps to get videos, then grab frames from them for ID? Or do you get both photos and videos?
Also, how big is your dataset? I.e., how many photos do you get of a particular butterfly species and of individual butterflies? 100s, 1000s?
I am asking because (1) IMO your dataset is quite valuable and could be published on its own; and (2) I wonder if you could automate the recognition of butterfly species and even individuals with some machine learning algorithm??? Is this something you’ve looked into??
Yes, I’m planning to publish after I analize all the data! (So probably end of this year hehe)
I use to get videos only because for some butterflies I need them to open the wings to fully id them. With the video I can see them moving around the flower and eventually they open the wings in the right position.
It’s a huge data set! I get a lot of videos, ~ 250-300 per day (9am to 4pm, 20 sec videos). When the weather is dark and rainy I get less. I have so many videos that I haven’t finish id them.
I would love some automate id algorithm but I haven’t look at it yet. Do you know something interesting that might work?
Yes, I’m planning to publish after I analize all the data! (So probably end of this year hehe)
Super excited to see this!
With the video I can see them moving around the flower and eventually they open the wings in the right position.
Makes sense.
I would love some automate id algorithm but I haven’t look at it yet. Do you know something interesting that might work?
With your large dataset, I suspect there might be a solution. I don’t know anything specific, but I peruse the AI for Conservation and Wildlabs communities, which have many people doing crazy AI stuff.
@deniseddWould you be able to share a few representative clips? If so, I can then reach out to those people and see what we can come up with!
Naqvi, Q., Wolff, P. J., Molano-Flores, B., & Sperry, J. H. (2022). Camera traps are an effective tool for monitoring insect–plant interactions. Ecology and Evolution, 12(6), e8962. https://doi.org/10.1002/ece3.8962
