Wild Edmonton: An automated image-processing AI solution for urban park cameras

The Problem

As urban expansion and infrastructure development continue to accelerate within the City of Edmonton, local natural habitats are shrinking, and animal movement patterns are rapidly shifting. To track these vital environmental changes and reinforce climate resilience, the City of Edmonton partnered with the University of Alberta to launch WildEdmonton. Since 2018, this initiative has deployed a network of over 109 remote imaging cameras across city parks, river valleys, and green spaces to document local wildlife.

However, the immense scale of data collection quickly triggered a severe logistical and ethical bottleneck. The network captures over 100,000 new images each quarter, swelling the historical dataset to more than 1.3 million images. Managing this inventory became a physically impossible task for the project team, who faced a massive backlog driven by "false triggers”: empty images caused by environmental factors like shifting vegetation or weather events. Compounding this operational strain were strict municipal privacy regulations. Because these public cameras frequently capture passing residents, privacy rules mandated that any image containing a person be completely scrubbed or pixelated before researchers, volunteers, or the public could access the data. Sifting through millions of photos manually to filter out false triggers and safeguard human privacy would have been a gruelling, financially prohibitive endeavour for municipal staff.

The Solution

To eliminate this bottleneck, the City of Edmonton’s Data Science and Research (DSR) team developed an automated, end-to-end image-processing infrastructure. The architecture operates through a clear division of data ingestion, algorithmic classification, and open cloud distribution.

Part 1: Automated Data Collection & Ingestion Pipeline The monitoring network relies on fieldwork execution, where camera equipment is strategically rotated and maintained quarterly across urban wildlife corridors. The resulting influx of 100,000 quarterly images is aggregated and funnelled into a central ingestion system handled by the DSR team.

Part 2: Computer Vision Classification & Privacy Redaction Engine Instead of relying on costly proprietary software, the team integrated Microsoft’s open-source CameraTraps Megadetector model. The DSR team customised and engineered a full software system around this foundational model, training it to autonomously detect and classify three core entities: humans, animals, and vehicles.

Part 3: Automated Extraction & Secure Cloud Delivery Once an image is processed, the system automatically filters out empty environmental false triggers. For images flagged as containing humans, the engine removes them from the main image pool. The remaining high-quality images are then automatically exported to a shared cloud drive. This creates a completely optimised data pipeline that allows University of Alberta volunteers and ecological researchers to bypass manual sorting entirely and focus 100% of their energy on tagging animal species.

The Impact

The deployment of the AI model shifted the WildEdmonton project from a stalled data backlog into an actionable urban planning asset. Key quantitative and qualitative outcomes include:

1. Drastic Operational Acceleration

The AI model analysed a staggering 800,000 images in a window of just two to three weeks—an analytical milestone that would have consumed hundreds of hours if performed manually.

2. Volunteer Efficiency Gain

By eliminating the burden of filtering empty frames and human interactions, the automated pipeline reduced volunteer species-tagging time by 62%.

3. Zero-Budget Technical Maturity

Developed entirely in-house by an internal team of eight using open-source tools, the solution required zero additional municipal budget allocation.

4. Tangible Urban Design & Safety Outcomes

By unlocking the dataset, city planners successfully identified narrow or fragmented ecological corridors. This intelligence directly informed the design of wildlife passages: corridors structured to allow safe animal migration through the city while minimising vehicular collisions and dangerous human-wildlife traffic interactions.

5. Expanded Ecological Discoveries

Seamless data processing empowered researchers to identify 28 unique mammal species and 45 bird species. This led to the discovery of two species entirely new to Edmonton's tracked boundaries (Elk and Raccoon), while verifying that invasive species like Wild Boar have not yet entered the municipality.

6. Civic Engagement & Public Trust

By utilising automated privacy redaction, Edmonton successfully balanced open-data transparency with individual privacy. The city can safely open-source wildlife imagery to neighbouring municipalities, partners, and the general public, empowering residents to become active environmental stewards.

Lessons Learned

• Prioritise Scrappy, Problem-Oriented Governance. Municipalities do not need multi-million-dollar software contracts to launch advanced AI solutions. By leveraging established, open-source models (like Microsoft's Megadetector) and utilising an in-house data science team, cities can build highly customised tools with zero direct software procurement costs.
• Embed Privacy by Design to Unlock Public Silos. Data security should not block municipal innovation. By building automated redaction directly into the ingestion phase of the workflow, public agencies can safely leverage data, maintain compliance, and seamlessly share findings with academic and regional partners.
• Leverage Multi-Sector Research Partnerships. The structural link between the City of Edmonton and the University of Alberta provides a highly transferable model. Connecting municipal data operations directly to academic talent ensures that local governments can tap into cutting-edge research while providing universities with massive, real-world datasets for high-impact community analysis.

Launch year: 2020

Privacy rights and legislation