simplifying wildlife research

​

Yorick Liefting

Wageningen University

13-04-2026

Camera traps have been around...

Shiras, G., July 1906. “Photographing Wild Game with Flashlight and Camera“, National Geographic, 17(7)

Camera trapping in the old days was cumbersome and yielded few images

1906, George Shiraz

1920, Frank Chapman

And have become affordable & widely used.

Potential for wildlife research

• Where do animals occur?
• In what abundance?
• What are trends in abundance?
• Interactions between species
• disease transmission (i.e. Swine Fever, Avian Influenza)
• Conservation & management questions
• ...

In principle, these questions can be answered (partially) with camera traps

Canis lupus, EOW Veluwe, Netherlands, Oct 2022

Published surveys

Unpublished surveys

Deployments by citizens

However…

Deleted data

Crashed hard drives

Data without evidence

Cherry-picked data

Some history

• Started in 2008 in Panama by Patrick Jansen & students (including myself)
• More projects came up...
• More images accumulated...

2016 - start of Agouti “2.0”

• Partnership between Wageningen University and the Research Institute Nature and Forest (INBO, Belgium)
• Towards a platform for all camera trappers
• Initially mostly used internally

Goals of Agouti.eu

Facilitate:

• managing camera-trap projects
• annotation & validation
• faster and standardized image processing
• bring “AI” to users, in an useful way
• long-term storage of images and data
• density estimation tools
• data publishing tools

While respecting data owners & the efforts they go though

The Agouti team

• Steering committee
• Responsible for the platform and overall strategy, finances, operation.
• Current members:

WUR

• Frank van Langevelde
• Yorick Liefting

INBO

• Jim Casaer
• Tanja Milotic
• Peter Desmet
• Tim Adriaens

​

• Development & Support
• Helpdesk agouti@wur.nl
• Data and project support
• Development
• IT operations
• Consisting of:
• 2 helpdesk staff
• 5 developers
• WUR IT department

As of 23 February 2026

​

Agouti hosts:

• 273.3 million images
• 142.6 thousand deployments
• 17.7 million observations
• 294.8 terabytes of images
• 3.441 species observed
• 1.853 projects
• 6.266 active users
• 369 organizations

​

New deployments per quarter

Data policy

• Data stored at WUR on campus data centers
• Data is owned by project owner/organization
• Images may be used internally for AI training
• Conditions of use:
• Complete deployments
• Intention to publish and share
• Commercial users and projects: provide support

​

Data standard: Camtrap DP

Massive amounts of camera trap data

Data management

Data processing

Data exchange, harmonization, publication

Suitable standards

​

Bubnicki et al. (2023) doi.org/10.32942/X2BC8J

Agouti

Wildlife Insights

TRAP PER

eMam mal

Use a simple data model

• This is hard!
• Three tables:
• Deployments
• Media
• Observations
• Supports wide range of:
• Deployment designs
• Classification techniques
• Analytical use cases

Develop openly

• Code on GitHub:
• Open source (MIT license)
• Versioned (incl. semantic)
• Collaboration on GitHub:
• Suggestions as issues
• Review, discuss, implement
• Automated tests
• Documentation on GitHub:
• Website (updates automatically)
• Example dataset

github.com/tdwg/camtrap-dp

Built into software

• Entire ecosystem:
• Management: Agouti, Trapper, other systems
• Analysis: Camtraptor
• Publication: GBIF IPT
• Challenging to coordinate
• Feedback from adopters

​

agouti.eu

os-conservation.org/projects/trapper inbo.github.io/camtraptor gbif.org/ipt

AI in Agouti

• Internal tools for Agouti.eu
• Contains:
• 5.2 million images with bounding boxes
• 800+ species (although heavily biased)
• Harvests data from agouti automatically
• Train custom models on selections of species/taxa
• Supports several architectures: edgeyolo, rt-detr, yolov9, convnext, etc.
• Model evaluation
• Main output: 100+ species model for Europe, currently at version 6.
• Underused and has potential. How to do that?

​

Collaboration:�European Observatory of Wildlife (EOW)

“To generate and provide information and unbiased trends on population abundance for those developing, adopting, implementing and evaluating environmental policy in Europe.”

• 40 countries, 65+ study sites
• One protocol, yearly surveys since 2022
• Uses the Agouti platform
• Open data, shared via the EOW consortium
• Link with policy: EFSA and Swine Fever

Collaboration:�WildObs

“The Wildlife Observatory of Australia (WildObs) is a national initiative building shared digital infrastructure to support the collection, processing, and reuse of wildlife camera-trap data across Australia.”

​

• Collaboration with Agouti
• WildObs uses the Agouti platform, but hosts the data within Australia
• Co-development of platform via shared codebase
• Luanch in June 2026
• https://www.wildobs.org.au

Collaboration:�CamerAI

• ‘Citizen science’ module for Agouti
• Interactive upload with interactive AI annotation
• Link with Observation.org
• Funded by Belgian Government
• Two pilot projects in summer 2026:
• Primary schools in Flanders
• Nature enthousiasts in their own gardens/property
• Live demo

​

​

2026: NLBIF project “Agouti FAIR”

Goal: Make Agouti camera‑trap projects FAIR* and lower barriers for publishing metadata/data via GBIF.

Funded by NLBIF

Key actions:

• New onboarding workflow for projects
• Agouti - GBIF integration
• Publishing workflow for projects (complete or metadata-only)
• Publication of WUR/INBO projects (lead by example)
• User campaigns to stimulate documentation and sharing

*Findable, accessible, interoperable, reproduceable

​

NLBIF Project

“Hedgehog Histories”

Ons doel:

• Data beschikbaar maken voor toekomstig onderzoek naar de achteruitgang van de egel in Nederland.

​

(Historische) data over:

• Datum
• Locatie
• Geslacht
• Leeftijd/gewicht
• Trekkers: Anouschka Hof & Bo Dillen

​

​

Tot nu toe:

• 9 opvangcentra van de 40
• Data van 2013 – 2025
• Meer dan 28.000 egels

​

Thanks for listening

​

Density estimation: Camera Calibration

• One-time process
• For each combination
• Camera model
• Resolution setting
• Camera parameters
• Centrally available in Agouti

​

Density estimation: Deployment�Calibration

• Calibration stick
• 1m
• 10 or 20 cm sections
• Manual registration of�calibration images
• Automated referencing�of pixels to ground level

Density estimation: Animal Tracking

• Camera model calibration
• Deployment calibration
• Tracking by observation
• Manual registration of animal�position on ground
• Recording the movement path
• Automatic calculation of
• Detection distance
• Detection angle
• Movement speed
• Input for the Random Encounter�Model (Rowcliffe et al. 2008)