Looking help with camera trapping for Jaguars: Software for species ID and database building

2021/02/23 21:49

Hi!

I'm looking for help with the development of a software for the analysis of thousands of photos from camera traps. We have more than 200 cameras deployed in field and the amount of information to store monthly is a lot. We have a long term monitoring project with camera traps since 2000.

We are in the stage that we need a software that is able to extract metadata from photos without an extra copy of each photo for a database but also be able to identify species. Internet connection is another challenge for us since the community we work doesn't have good internet access.

Any help is greatly appreciated,

Carmina

Edited on Fri, 2021-02-26 13:37 by Carmina Gutierrez, Updated thread name to accurately reflect ask.

juan

2021/02/26 14:50

Hi Carmina,

That sounds like a massive effort, and you do need to be very organized otherwise things can get out of hand.

Could you clarify whether you are looking to develop your own software or use the ones available? There are several softwares now that follow the same general standard. Young et al. did a review of some of the most popular ones (see pdf attached).

I've had good experience with camtrapR, which is an R package. It has great flexibility in terms of setting up a personalized workflow, but it's definitely not user-friendly, especially if you are working with community volunteers.

The other softwares are fairly similar, I've used Wild.ID and Camelot. I especially like Camelot, it's very well built, and there is a good user community where the developers actively help. These are very good at keeping things organized for long-term surveys in my opinion, you can keep track of individual cameras (identified by serial number or unique code), different sites, and different surveys. Camelot has the added benefit of automatic detection of animals. It doesn't identify species, but it can detect which photos have animals, people, or vehicles. You do require a good connection for this, but all the other functions don't. The downside would be that it does create a separate copy of your images, which you said you didn't want to.

Wild.ID is also still active, and it has been integrated with Wildlife Insights. It keeps the same overall workflow and structure as Camelot, but I think you don't have to create another copy of the photos.

I hope this helps.

Best of luck,

Juan

Carmina Gutierrez

2021/02/26 18:31

In reply to juan

Hi!

Thanks a lot for your suggestions, I was thinking in trying Camelot, as you said, I've looking for options that can help us. The reason we don't want an extra copy of the photos associated to a database is because we have so many photos each mont/year that this will simply be too much for any program and computer. I saw in the paper that their capacity is usually around 100,000 photos or no more than 1 million the best ones.

Wildlife insights is exactly what we need but the internet is our problem. We have a terrible internet connection that makes very difficult to upload photos. We had a volunteer that was creating a sofware for us able to extract metadata and identify some species without an extra copy of the photos and that didn't need internet (run in the computer with the training of the most common species) but he suddenly stopped helping us (I think because of pandemics). We are looking for someone willing to do something similar.

We have like 30 common species in the cameras, that would be more what we need to be identified or even something like what Camelot does (animal, people, vehicle, nothing) could help and would reduce our time a lot.

Suggestions are welcome!

Thanks!

Carmina

juan

2021/03/02 15:44

In reply to Carmina Gutierrez

Hi Carmina,

I definitely get not wanting to create another copy of your data, I realized this about Camelot just a little late, and we did have to buy a 4 TB external hard drive. The unique names that Camelot assigns to its copies of the data are also not friendly if you want to move things around.

I mentioned that Wild.ID is being more and more integrated with Wildlife Insights, but it can still be used as a standalone software. It has the same format as Camelot, but I think it doesn't create and additional copy (I might be wrong about this, though). You should take a look at this, it might be what you need.

As for detection, Camelot puts together batches of photos, uploads them to Microsoft's servers, and then gets the resulting database back. The model they use, however, is available and you can run it on your own computer. The AI model that it uses is called Megadetector. You used to have to run it from the command line, but recently Petar Gyurov posted this GUI that makes using Megadetector soooo much easier. You just tell the software where your photos are, and let it run. You can decide whether you trust it enough to sort your photos automatically, or check yourself. In most conditions it performs very well, you may have issues when you can only see parts of an animal behind some vegetation. The check would take less time than going through the raw pictures, since it marks boxes where it found things. You still have to identify the pictures after they're sorted. Megadetector will work on any computer, but it performs much faster if it has an Nvidia graphical card. However, setting this up is a little tricky.

Finally, the output from Megadetector can be integrated into the workflow of Timelapse, another software. I've never used this one so I can't really say how well it works.

Integrating everything can be a little intimidating, so feel free to ask if you need guidance for any of these steps.

Best,

Juan