Gecko monitoring: Choosing the right software to analyse Wildlife Watcher photos

Introduction

Camera traps have become invaluable in wildlife conservation, providing a non-invasive way to monitor animal populations, track behaviours, and study ecosystems. Having a way to capture images and videos of wildlife in their natural habitats offers more in-depth insights than direct observations.

However, the vast amounts of data collected could be time-consuming to process. Manually sorting thousands of images to identify species and remove empty images (false triggers) is a task many camera trap users face at some point.

Thankfully, nowadays, numerous camera trap software solutions exist to process the footage. From open-source tools to commercial platforms, these applications offer features such as species identification, data management, or project management. Choosing the right tool mostly depends on the needs of the camera trap project.

As developers and early users of the Wildlife Watchers, we have specific requirements. If you don’t know what our Wildlife Watchers are, please read the Wildlife Watcher project description. In short, the Wildlife Watchers are cameras that can record any moving animals and use AI to identify them. To produce the Wildlife Watchers, we’ve outlined two Minimum Viable Products (MVP) tailored to specific use cases:

1. MVP 1 focuses on monitoring threatened New Zealand geckos. The device can capture a wide range of wildlife, including reptiles and insects, in various weather and lighting conditions.
2. MVP 2 focuses on predator detection, specifically identifying invasive rodents. This version includes species identification, wireless communication, long-lasting battery life, and an easy-to-use mobile app.

In this post, we will focus on the requirements for MVP#1, we will have a separate post for the requirements for MVP#2.

The user requirements for the Wildlife Watcher MVP#1 are:

• A user-friendly mobile app for deploying and retrieving devices.
• A reliable sensor that activates the camera when a species of interest is present.
• High-quality images that enable differentiation between similar-looking species (e.g., geckos vs. skinks).
• A streamlined process for converting images into information about animal occurrences and absences.

With these requirements in mind, what are the best available software options for camera trappers? Are there species-specific solutions?

To answer these questions, we evaluated eight software solutions across three categories: desktop applications, web applications, and software packages. This comparative analysis is based on our team’s experience and commonly used tools. While not an exhaustive review, we welcome suggestions for other solutions we may have overlooked.

What’s available?

Web Applications

1. Agouti

Agouti is a wildlife monitoring web platform designed for managing and analysing camera trap data.

Strengths:

• Open source and free.
• Features an efficient workflow and comprehensive documentation.
• Designed for processing large volumes of camera trap images, which is more efficient for wildlife research.
• Retains and processes important camera trap metadata, such as timestamps, GPS locations, and environmental conditions
• Includes GIS-based tools to analyse species distribution, movement patterns, and habitat use. Great for ecological researchers.
• It’s free for Non-Profit Organizations (like Wildlife.ai).
• Supports collaboration (multiple users to work on datasets).

Limitations:

• Limited for underrepresented animals of most camera trap data (small mammals, cold blooded, insects…).
• Feature-rich platform but complex, requiring some training to use effectively.

2. Wildlife Insights

Wildlife Insights is a web platform for managing and analysing camera trap data with a focus on conservation.

Strengths:

• Uses Google’s open source machine learning models (SpeciesNet, we will talk about it later on) to automatically identify species.
• Supports bulk uploads and processes images efficiently, which is ideal for large research projects.
• Metadata management, the user can add information about the deployments, cameras, and species found.
• Collaboration & Multi-User Access is here and requires an individual account per user and project creation for the project manager.
• Integration with GBIF & Conservation Networks – Wildlife Insights allows data to be shared with global biodiversity databases, supporting conservation efforts.

Limitations:

• Limited free usage.
• Limited for underrepresented animals of most camera trap data (small mammals, cold-blooded, insects…)
• Requires a stable internet connection, which can be a challenge in remote fieldwork settings.
• Structured workflow that may require training for researchers unfamiliar with large-scale camera trap data management but without technical expertise.

3. eMammal

eMammal is a data management system and archive for camera trap research projects. 

Strengths:

• Open source and free.
• Both researchers and citizen scientists can help classify species.
• Automatically extracts and organises camera trap metadata, including timestamps, locations, and environmental conditions.
• Uses a scientific protocol for data collection, perfect for consistency across projects and large-scale comparisons.

Limitations:

• Focusing on mammals, limited capabilities for underrepresented animals in common camera trap data (small mammals, cold-blooded, insects…)
• Not optimised for Bulk Processing.
• Manual Species Identification Required.
• Need to ask for project creation.
• Strict guidelines, making it less flexible.

4. TRAPPER

Trapper is an open-source tool for managing and analysing camera trap data.

Strengths:

• Open source and free.
• Supports bulk image imports and processing.
• Reads and organises camera trap metadata.
• Works with relational databases (e.g., PostgreSQL), great for data management.
• Offers both command-line tools (great for automation) and a graphical interface.
• Offers installation and configuration guides, open-source code, and multiple support resources to assist with deployment and usage.

Limitations:

• Requires a complex initial setup with software installation, database management, and command-line operations
• Limited capabilities for underrepresented animals in common camera trap data (small mammals, cold-blooded, insects…).

5. iNaturalist

iNaturalist is a popular citizen science platform designed for wildlife observations and species identification.

Strengths:

• Open source and free.
• Crowdsourced and AI Species Identification.
• Global Community Engagement thanks to a widely-used platform.
• iNaturalist follows standardised taxonomy and integrates with databases like GBIF (Global Biodiversity Information Facility).
• Got a species recognition model that identifies underrepresented animals in common camera trap data (small mammals, cold-blooded, insects…).
• Usable on the web or a phone application, but also via their application programming interface (API).
• The Machine Learning behind iNaturalist could be integrated into a custom application if needed.

Limitations:

• Not specifically built for camera trap data.
• iNaturalist is optimised for individual observations rather than large-scale camera trap datasets.
• Limited support for detailed metadata beyond GPS and time.
• Requires a complex initial setup with software installation, database management, and command-line operations.
• By default it’s not designed for managing camera trap data.

Desktop Applications

6. Wild ID

Wild ID is an online detection and identification of Southern African wildlife species in camera trap images using machine learning software.

Strengths:

• Open source and free.
• Supports bulk image imports and processing.
• Runs locally on a server or computer, good for sensitive data and areas with limited internet access.
• Works with relational databases (e.g., PostgreSQL).
• Offers both command-line tools (great for automation) and a graphical interface.

Limitations:

• It requires registration, the installation process requires a medium level of technical expertise, and is specialised for Southern African wildlife species.
• Reads and organises only basic camera trap metadata (timestamps, GPS locations, and camera settings).
• Limited capabilities for underrepresented animals in common camera trap data (small mammals, cold-blooded, insects…).

7. Camelot

Camelot is an open-source camera trapping software for wildlife researchers and conservationists.

Strengths:

• Open source and free.
• Supports batch imports, tagging, and filtering.
• Automatically reads and organises EXIF metadata from images, including timestamps, GPS coordinates, and camera settings.
• Deployment could be local or server-based.

Limitations:

• It requires a specific Java distribution (Temurin), connects to the internet at launch, and is mainly for live video camera trapping.
• Setting up the application requires an advanced level of technical expertise.
• Requiring manual classification.

Software Packages

8. camtrapR

CamtrapR is a local programming language (R) package for processing and analysing camera trap data.

Strengths:

• Open source and free.
• Built specifically for wildlife research and its data analysis workflows.
• Usable with R and its advanced analysis tools (creating camera trap maps, effort maps, or species distribution maps).
• Runs locally on a server or computer, perfect to have control over sensitive data or run the application on remote locations.

Limitations:

• The tool is limited to R users and requires integration into a R-based web-application or an R local software.
• No Built-In AI Species Recognition.
• Since it runs in R, camtrapR lacks a graphical user interface.

9. SpeciesNet

Newest local programming language (Python) package for processing and analysing camera trap data.

Strengths:

• Open source and free.
• Uses a famous object detection (MegaDetector) and then recognise the cropped image of the animal, which increases the precision of the model.
• Trained on 65M images and 2000 labels.
• Detailed documentation.

Limitations:

• Reads and organises only basic camera trap metadata (timestamps, GPS locations, and camera settings).
• Does not offer metadata organisation or data management functions.
• Tool is limited to Python users and requires integration into a Python-based web-application or a Python local software.
• Requires Technical Expertise.
• Requires Cloud or High-Performance Infrastructure.

Results

Our evaluation of popular camera trap software solutions identified several options with potential for the Wildlife Watcher project. Agouti and Wildlife Insights emerged as the leading candidates for comprehensive camera trap data management. These tools offer project and deployment management, efficient annotation features, user-friendly interfaces, and free usage for NGOs.

While Agouti and Wildlife Insights provide robust data management capabilities, iNaturalist and SpeciesNet stand out for their species recognition accuracy. However, iNaturalist is not designed for camera trap data management and SpeciesNet requires integration into a Python or web-based solution to fully meet our needs.

Several tools were deemed unsuitable for our project due to limitations in functionality or complexity. eMammal’s focus on mammals does not align with our broader conservation goals. Trapper, WildID, Camelot, and camptrapR were considered too complex for our target users, who may not have extensive technical expertise.

Conclusion

While a perfect, all-in-one solution remains abstract, our evaluation has identified a few promising tools for the Wildlife Watcher project.

Agouti and Wildlife Insights stand out for their comprehensive data management features and user-friendly interfaces. We will be actively testing and utilising these platforms in our upcoming projects to experience their capabilities firsthand.

We have to mention the better-suited species recognition capabilities of iNaturalist and SpeciesNet. Knowing that SpeciesNet is the model behind Wildlife Insights, Wildlife Insights seems to offer both robust data management and advanced AI-powered species identification. We still need to test how the model identifies underrepresented animals in common camera trap data (small mammals, cold-blooded, insects…).

And as we move forward, our focus will be to test and integrate Wildlife Insights into the Wildlife Watcher project and our specific conservation needs. We will be posting here how the system works so stay put!

Key Takeaways

• Best for AI-Driven Identification: SpeciesNet, Wildlife Insights, iNaturalist (crowdsourced identification).
• Best for Large-Scale, Long-Term Projects: Agouti, Wildlife Insights, eMammal.
• Best for Customisation: Agouti, Trapper, camtrapR.
• Best for Offline, Local Use: Trapper, Wild.ID, Camelot.
• Best for Collaboration: Agouti, Wildlife Insights, eMammal.

Appendix: Detailed pros and cons for our case scenario