Would also add zamba and Zamba Cloud into the list of tools for running AI models (the former being with Python code, the latter being with a GUI). For the reader at home (since I know Dan knows), these are tools for species classification on camera trap videos. We've got pretrained models trained on over 250,00 videos covering 32 African forest species, 11 European species, as well as a blank / nonblank detector.

There is also the ability to train your own models for new species or locations. And we're super excited to extend this custom model training functionality to support camera trap images thanks to our WILDLABS award (more details here).

Thanks Dan for explaining and clarifying, and Ed and all others for contributing. As a somewhat technical person (IT business consultant) with no formal environmental education, but with a passion to understand and protect the environment, this is a topic I've been exploring and struggling with...

There are really skilled technical experts in this community that love to design and improve solutions, but it's not always easy for potential end users to understand and use said solutions without significant learning curves. And often solutions are designed from a technical perspective, and does not always consider end users' level of experience, knowledge and actual on the ground needs. This is not just a problem in the concervation technology space, but overall tech industry.

To clarify this for myself, I would love to see a summary of how conservation tech solutions fit into a camera trap process that include the following areas - and especially which tools fit where:

1. Plan/Design: scope, requirements, timeframes, data needs, overall outcome. Also influences the selection of tools to use in the overall process.
2. Capture/execute : both hardware (cameras, audio recorders, sensors, etc), and software (managing hardware lifecycle tracking, capturing trail camera data e.g. image file management, etc) solutions
3. Process/Analyse: solutions for initial processing/culling, detailed analysis, and anything classification related
4. Present/Display: how best to communicate, display and present the data gathered, analysed and classified...

If there is a resource that already do this comparison and sumarization of tools' capabilities please help me find it - otherwise it might be topic to be explored further...

Saul Greenberg is a great guy! He has made a few very useful videos regarding the Timelapse/Megadetector integration:

https://grouplab.cpsc.ucalgary.ca/grouplab/uploads/Publications/Publications/2021-05-ImageRecognition-Video.mp4

and a tutorial: 

https://saul.cpsc.ucalgary.ca/timelapse/uploads/Videos/Video-TimelapseImageRecognitionGuide.mp4

Happy to explain for sure. By Timelapse I mean images taken every 15 minutes, and sometimes the same seals (anywhere from 1 to 70 individuals) were in the image for many consecutive images. 

I took delivery of the DJI Mavic Enterprise 3 Thermal the other day. The short hand nomenclature used on the controller for the thermal imagery is "IR" (short for infrared) so it is used even for cases where no IR lighting is in play.

After reading through the backlog of comments, I had a thought triggered by the mention of security cameras. Some security cameras don't get triggered by the usual PIR sensor, but by changes in the image in pre-set zones. I have a Bosch Binion starlight camera I was playing with and it worked this way. The area of the trigger zones could be adjusted once you had the camera set up and viewing the monitoring area. It wouldn't matter how slow the sloth moved, when it moved into the zone and changed the image, the camera would trigger. I assume sensitivity could be adjusted so you didn't get a trigger every time a leaf moved in the wind.

Thanks Dan! I did actually, after giving up using PyTorch and it was amazing!

This is great, thank you so much @zhongqimiao ! I will check it out and looking forward for the upcoming tutorial!

I doubt there is an off the shelf solution. Likely you will have to build one. Again I think the FLIR leptons could be of value here.

@krasi_georgiev  you have worked with Leptons before ? Is this something you are able to advise on ?

@mguins should share some more details on the project. IIUC she needs IR at night and not thermal images so a simple CCTV camera with some code will do the job.

Ah yes. I didn’t notice that. Indeed it’s “near infrared”, 850nm lighting.

Thanks for the update @Frank_van_der_Most . I have been curious about the AA Li-Ion/Li-Po batteries and how they perform. The sudden drop in reported voltage will likely be from the internal voltage regulator switching off when the internal cell gets to the low voltage threshold (usually around 3V) to avoid damage to the cell. Looking now at a discharge curve they show a constant 1.5V, then a step down to 1.1V before dropping to zero. I don't know how you can possibly test these externally to know how much energy they have left until you hit the 1.1V step. 

Hi Alistair @alsnothome , see the comments of @Amanda_Matthes at the post here. Basically, it is not easy

Hi Frank,

Yes, I agree.  There is a halfway-house solution if you take a look at the Energizer Ultimate Lithium range of batteries.  They have superior life to alkaline batteries (3.5Ah in a single AA cell), though they are of course still one-time use.  They would also work a little better with your battery monitor since they have a graceful degradation between 1.8V down to 1.5V, but then they fall off a cliff ;-)

A multi meter would probably not be enough, unless you have a very fancy one. To get the energy use, you need to be able to integrate the current drawn over time.  Something like this: Otii Arc Pro

I don't know how much you paid for these, but Amazon Basics has a line of rechargeable AA batteries, including a high capacity version which can store 2400 mAh which is a little more than the ones you're using.  (I've seen even more capacity with other manufacturers.)

You might also want to consider avoiding batteries with USB ports in the future. It seems to me like just an additional thing that can break, especially if moisture could be an issue. 

@Lars_Holst_Hansen  Here is the information you requested. Also run Yolov8 in multiple remote environments without any issues.  Perhaps you'll need to use a virtual environment (venv et al) or conda to remedy incompatibility issues. 

$ yolo checks
Ultralytics YOLOv8.1.4 🚀 Python-3.10.12 torch-1.13.1+cu117 CUDA:0 (Quadro T2000, 3904MiB)
Setup complete ✅ (16 CPUs, 62.5 GB RAM, 465.0/467.9 GB disk)

OS                  Linux-6.5.0-17-generic-x86_64-with-glibc2.35
Environment         Linux
Python              3.10.12
Install             pip
RAM                 62.54 GB
CPU                 Intel Core(TM) i7-10875H 2.30GHz
CUDA                11.7

matplotlib          ✅ 3.5.1>=3.3.0
numpy               ✅ 1.26.3>=1.22.2
opencv-python       ✅ 4.7.0.72>=4.6.0
pillow              ✅ 10.2.0>=7.1.2
pyyaml              ✅ 6.0.1>=5.3.1
requests            ✅ 2.31.0>=2.23.0
scipy               ✅ 1.11.4>=1.4.1
torch               ✅ 1.13.1>=1.8.0
torchvision         ✅ 0.14.1>=0.9.0
tqdm                ✅ 4.66.1>=4.64.0
psutil              ✅ 5.9.8
py-cpuinfo          ✅ 9.0.0
thop                ✅ 0.1.1-2209072238>=0.1.1
pandas              ✅ 1.5.3>=1.1.4
seaborn             ✅ 0.12.2>=0.11.0