Hi all! Folks may be interested in the Cornell Tree Climbing program that is a part of Cornell Outdoor Education. Not only does Cornell offer training, and have a bunch of online resources, but they have also facilitated groups of scientists to collect canopy samples and data. 

Read "Cornell Tree Climbing | Student & Campus Life | Cornell University" on Student & Campus Life | Cornell University

Cornell Tree Climbing | Student & Campus Life | Cornell University

CTC promotes safe and environmentally responsible tree climbing techniques for

Student & Campus Life | Cornell University

Hello Carly, 

Congratulations for this project!

I am studying right now a second MA in Environment Management. I would like to do my MA thesis project about these technologies (bioacoustics and camera traps). I wonder if you would be interested in a collaboration with me on that ? 

I already have a professional experience so I think that my contribution could be interesting for you. 

Thank you in advance for answering, 

Simon

A good candidate for low power hibernation and processing power is the Teensy 4.1 from PJRC, which is an ARM Cortex M7. standard clock is 600 MHz and there is provision for 16 MB PSRAM. It hibernates at 0.1 ma. What is more important than the specs, there is a very active (the best IMHO) community Search | Teensy Forum with direct connection to the developer (Paul Stoffregen). For a simple recorder consumption is 50% higher than RP2040 (Teensy running at 24 MHz and RP2040 running at 48 MHz, but RP2040 is M0+ and not a M7). 

Hi Raza,

I manage a lot of snow leopard camera trap data also (I work for the Snow Leopard Trust). I am not aware of any AI solutions for snow leopards that can do all the work for us, but there are 'human-in-the-loop' approaches, in which AI suggests possible matches and a human makes the actual decision. Among these, are Hotspotter and the algorithm in WildMe. 

You can make use of Hotspotter in the software TrapTagger. I have found this software to be very useful when identifying snow leopards. Anecdotally, I think it improves the accuracy of snow leopard identifications. But, like I said, you still have to manually make the decisions; the results from Hotspotter are just a helpful guide.

The other cutting edge solutions mentioned here (e.g. MegaDescriptor, linked above) will require a massive dataset of labelled individuals. And considerable expertise in Python to develop your own solution. I had a quick look at the paper at they were getting around 60-70% accuracy for leopards, which is a much easier species than snow leopard. So I don't think this approach is useful, at least for now. Unless I've misunderstood something (others who deeply understand this, please chime in and correct me!).

Incidentally, I did try to gain access to WildMe / Whiskerbook last year but wasn't successful gaining an account. @holmbergius can you help me out? That would be appreciated, thanks!

Best of luck Raza, let me know if I can help more,

Ollie

An example of Hotspotter doing quite a good job, even with blurry images, to successfully draw attention to matching parts of the body. This is a screenshot from TrapTagger, which very helpfully runs Hotspotter across your images.

Note that I wouldn't necessarily recommend ID'ing snow leopards from such poor imagery, but this just demonstrates that Hotspotter can sometimes do quite well even on harder images.

Hi Raza,

As @ollie_wearn suggests, if think traptagger will be the easiest for you: 

Read " Sign In - TrapTagger " on

Sign In - TrapTagger

You just have to follow the tutorial there:

Read "TrapTagger Tutorial - YouTube" on YouTube

TrapTagger Tutorial - YouTube

The full TrapTagger tutorial. It is recommended that you start here, and watch through all the videos in order. You can then revisit topics as needs be.

YouTube

The person in charge of Traptagger is also very responsive.

best,

Sam  any update on Instant Detect 2.0 - previously you mentioned that you hope to go into volume production by mid-2024?

I would love to also see a comparison between Instant Detect 2.0 and Conservationxlabs' Sentinel products if anyone has done comparisons.

Are there any other similar solutions currently on the market - specifically with the images over LoRa capability, and camera to satellite solution?

As others have said, pretty much all image models at least start with general-subject datasets ("car," "bird," "person", etc.) and have to be refined to work with more precision ("deer," "antelope"). A necessity for such refinement is a decent amount of labeled data ("a decent amount" has become smaller every year, but still has to cover the range of angles, lighting, and obstructions, etc.). The particular architecture (yolo, imagenet, etc.) has some effect on accuracy, amount of training data, etc., but is not crucial to choose early; you'll develop a pipeline that allows you to retrain with different parameters and (if you are careful) different architectures.

You can browse many available datasets and models at huggingface.co 

You mention edge, so it's worth mentioning that different architectures have different memory requirements and then, within the architecture, there will generally be a series of models ranging from lower to higher memory requirements. You might want to use a larger model during initial development (since you will see success faster) but don't do it too long or you might misestimate performance on an edge-appropriate model.

In terms of edge devices, there is a very real capacity-to-cost trade-off. Arduinos are very inexpensive, but are very underpowered even relative to, e.g., Raspberry Pis. The next step are small dedicated coprocessors such as the Coral line (). Then you have the Jetson line from NVidia, which are extremely capable, but are priced more towards manufacturing/industrial use. 

Read "Products | Coral" on Coral

Products | Coral

Helping you bring local AI to applications from prototype to production

Coral

Regarding using the openMV as a basis for an allrounder camera, we just published this:

I just noticed that TrapTagger has integrated AI reading of timestamps for videos. I haven't had a chance to try it out yet, but it sounds promising. 

Hi Lucy,



I now realised it is an old thread and you most likely have already found a solution long ago but this might be of interest to others.

As mentioned previously, it is definitely much better to take moon phase from the date and location. While moon phase in general is not a good proxy for illumination, that moon phase symbol on the video is even worse as it generalises the moon cycle into a few discreet categories. For calculating moon phase you can use suncalc package in R but if you want a deeper look and more detailed proxy for moonlight intensity, I wrote a paper on it 

Read "Biologically meaningful moonlight measures and their application in ecological research | Behavioral Ecology and Sociobiology " on SpringerLink

Biologically meaningful moonlight measures and their application in ecological research | Behavioral Ecology and Sociobiology

Light availability is one of the key drivers of animal activity, and moonlight is the brightest source of natural light at night. Moon phase is commonly us

SpringerLink

 with accompanying R package called moonlit 

Read "GitHub - msmielak/moonlit: moonlit - R package to estimate moonlight intensity for any given place and time" on GitHub

GitHub - msmielak/moonlit: moonlit - R package to estimate moonlight intensity for any given place and time

moonlit - R package to estimate moonlight intensity for any given place and time - msmielak/moonlit

GitHub

When it comes to temperature I also agree that what is recorded in the camera is often very inconsistent so unless you have multiple cameras to average your measurements you are probably better off using something like NCEP/NCAR Reanalysis (again, there is an R package for that) but if you insist on extracting temperature from the picture, I tried it using tesseract and wrote a description here: 

Read "Michał Śmielak - Extracting date and time from camera trap photos using R and tesseract" on

Michał Śmielak - Extracting date and time from camera trap photos using R and tesseract

Good luck!