Wildlife Camera Traps - of the future

picamera

#1

I am pretty sure that wildlife camera traps are significantly lagging, technologically.

Am picturing all sorts of Machine Vision and inter-device communication, bringing intelligence to the field…Pi or Fin boards with relatively inexpensive, but TPZ cameras…the network making decisions about which animals or birds are the most “interesting” from a rarity or other perspective…

I have a friend with 200 acres in a wildlife corridor in Costa Rica, the Paseo de la Danta, above Uvita…there have been jaguar sightings, and we found a puma print on the ridge that overlooks the Whale’s Tail…I would really like to help big cat research.

As an Analyst, I understand the basic concepts of thin clients/IoT and Docker. This video seems to pretty well describe one viable hardware setup.

In dialog with Jordan at resin: “…could run image classifying on the Fin, but more-specialized hardware might be able to do it faster or more efficiently, like one of the nVidia boards. Alternatively, the image processing could be done off the device, with the device used for camera management and storage before offloading to something like AWS Rekognition.”

How would I best go about developing an Open Source Model project?


#4

Hi @kenlyle ,
thanks for posting this on balena forums, sounds like an interesting idea.
Are you looking at implementing this on balena ?
Cheers Thomas


#5

Yeah, I was thinking Cloud.


#9

As Jordan suggested, there are a few ways to go about developing some image recognition software - there are software as a service platforms that could do this for you (such as AWS Rekognition as suggested), or you could get into the process of doing everything at the edge with a TX2. That decision I guess would depend on your connectivity in this particular situation - you’re not going to be able to stream images and video to AWS for processing if your cameras are out in the wilderness with no internet access.

I found this article which along with the linked study seems quite relevant: https://medium.com/coinmonks/automated-animal-identification-using-deep-learning-techniques-41039f2a994d


#13

And maybe this and/or this


#14

Thanks, Chris. Great article.

Amazon offers this package of the TX2:

Would a preliminary architecture for the remote site be something like the above, a Wi-Fi router, a dozen PTZ cameras and power, weather protection, and mounts for all?

We do have signal in the area from at least one cellular provider, as well.

I am understandng that, if we do it right, the idea is that we filter the empty images, etc. locally, and only send the good stuff to storage, which will reduce costs.

I am thinking that we might be able to make the project self-funding by monetizing the live video content by setting up server-side ad insertion using AWS Elemental MediaLive, AWS Elemental MediaPackage, AWS Elemental MediaTailor, and Amazon CloudFront on some kind of membership website.