Hello everyone,
I’ve been running a set of trail cameras for several months and now have thousands of images showing consistent activity from what appear to be 1–2 groups of deer, a local black bear family with 2-3 cubs, and at least one family of bobcats with 2 kittens.
My goal is to go beyond just species identification and start tracking individual animals over time, essentially, building a catalog so I can better understand their movements, interactions, and activity patterns.
Here’s what I’d like to do:
- Sort images by species automatically (deer, bear, bobcat).
- Recognize individuals, particularly bobcats (spot patterns are unique), possibly bears (chest/face markings), and deer (antler shape, scars, seasonal coats).
- Monitor activity by cataloging sightings across dates/times/locations.
I’d love to hear from anyone who has experience with:
- Practical workflows for moving from raw trail cam images → species ID → individual ID.
- Software that is easiest to set up for someone without a full ML engineering background.
- Tips for labeling images and building a reliable monitoring catalog.
- Any open-source projects or collaborations that might welcome additional datasets like mine.
My ideal outcome would be a repeatable pipeline where I can upload new batches of images and get updated insights into which individual animals are returning and how often.
Thanks in advance for any advice, tools, or guidance you can share!
9 September 2025 8:02pm
Hey! I'm pretty new to machine learning, but I can offer some insights based on the project I've worked on.
Data Collection Annotation
- Data collection and proper data annotation were easily the most time consuming but meaningful steps in terms of improving model performance
- I was able to train my model in 1/8th of the time with a smaller customized data set that was using images from the environment my camera was actually operating in as opposed to using an open source data set
- Data collection and annotation strategies, tips, and tools for can be found in this guide:
CV Data Collection and Annotation Guide | Ultralytics
Data collection and annotation are vital steps in any computer vision project. Explore the tools, techniques, and best practices for collecting and annotating data.
- Annotation Tips: bounding boxes were the easiest to start off with, identify rules for how you are going to annotate for consistency, verify your annotations and dataset frequently (visualization was an easy way for me to find outliers or issues)
- I wonder if you could also use an existing ML model to take an initial stab at classifying or bounding your images at least. Just stumbled upon this tool as well, could be helpful:
- Tools: many annotation tools are referenced in the link above, but roboflow is another solid place for setting up workflows, finding data sets and CV/ML resources in general
I have no experience with identifying individuals, but seems like you hierarchical classification would be helpful? Where based on confidence levels or other parameters and strategies you classify as either the species or individual.
If you are starting out with ML the YOLO guides have really helped me. I hope this helps, good luck!
20 September 2025 4:50am
I work on the BearID Project, developing models for individual bear ID (mostly brown bears so far). I also collaborate with TRAPPER, which is an open source camera trap management package with AI pipelines. We have a WILDLABS funded project called Trapper Keeper to try to make all this easier. Check it out!
Anam Khan