Hyland Lake Park Reserve Deployment
A 14-day observational pilot designed to test the end-to-end feasibility of deploying solar-powered, cellular-enabled trail cameras in a real park environment, conducted in partnership with Three Rivers Park District.
Deployment Overview
Deployment Parameters
Study Objectives
- •Test operational viability of solar-powered, cellular-enabled cameras
- •Evaluate data throughput and transmission reliability
- •Assess preliminary detection quality
- •Demonstrate platform analytics capabilities
- •Identify areas for AI algorithm improvement
Camera Placement
Seven solar-powered, cellular-enabled cameras were deployed across strategic locations: 3 cameras at park entrance points, 2 cameras on dedicated ski trails, and 1 camera at a key crossing/intersection point. All units were installed in tree canopy via ladder at approximately 10 feet mounting height.
Deployment Results
Note: Vehicle (car) detections were excluded from this analysis to focus on pedestrian and trail-user activity.
Platform Capabilities Demonstrated
Real-time analytics dashboard showing detection patterns across all stations with classification breakdown
Time-Series Analysis
Full time-series based charts supporting sample windows from 15-minute intervals up to 1 month. Bar/line style charts with ability to combine stations or classes.
Category Comparison
Ability to analyze time-history comparing site data and classifications across sites.
Quick Auditability
Click data points to see a sample of the actual detection image for quick validation.
Data Reclassification
Ability to customize data classification, relabel classifications, and map classes with a many-to-many capability.
Site Performance
6 of 7 Sites Performed Flawlessly
86% SuccessSites 2–7 transmitted data consistently throughout the entire 14-day observation window without interruption. One site (Site 1) did not transmit due to an installation error, which has since been identified and resolved for future deployments.
Solar/Power Performance
Despite winter light conditions and high frequency detections, no camera ran completely out of batteries throughout the 14-day period. However, we observed snow accumulation on one solar panel after heavy snowfall, which limited recharging. This finding highlights the importance of installation methodology and suggests solar panels may benefit from periodic brushing or angled positioning in winter conditions.
Snow buildup observed
Algorithm Evolution: From Pilot to Production
This pilot played a critical role in the development of our detection algorithms. The diverse and challenging winter conditions at Hyland Lake provided invaluable real-world data that directly informed our AI improvements.
VisionAI v2: Built on Real-World Insights
The challenging conditions encountered during this pilot—winter lighting, snow accumulation, shadows, and high-frequency motion triggers—helped us develop VisionAI v2. Our current algorithm achieves above 96% subject detection accuracy across diverse and challenging environments, incorporating advanced features like close-proximity unique subject filtering (CPUSF) and adaptive confidence thresholds that were directly informed by this study.
Key Learnings from the Pilot
Every field deployment yields engineering and methodological insights. The following observations emerged from this study and will inform future instrumentation, placement, and analytical approaches.
1. Environmental Variables Are A Factor
Despite winter light conditions and high frequency detections, no camera ran completely out of batteries throughout the 14-day period. However, snow did accumulate after a heavy snowfall which limited recharging on a few cameras.
2. Real-World Data Drives AI Innovation
The diverse conditions captured during this pilot directly informed the development of VisionAI v2. Insights from this study led to advanced features like close-proximity unique subject filtering (CPUSF) and adaptive confidence thresholds, enabling our current 96%+ detection accuracy.
3. Proper Installation Training Ensures Success
6 of the 7 cameras worked flawlessly throughout the study, with one site experiencing an installation error that has since been resolved. This highlights the value of providing land managers with proper training on device installation and solar panel setup to ensure consistent results across all deployment sites.
Next Steps & Areas of Interest
The Hyland Lake pilot demonstrated operational feasibility and surfaced meaningful questions for further investigation. The following areas represent natural next steps—presented here as open directions for collaborative discussion rather than a fixed roadmap.
Formal Validation Study
A structured accuracy assessment comparing vision-based detections against manual ground-truth counts. Ideally conducted in partnership with an experienced agency such as Three Rivers to ensure methodological rigor and practical relevance.
AI Directionality Modeling
Development of AI-based estimation of subject travel direction from imagery sequences. This capability could enable origin-destination analysis on trail networks—a significant advancement over bidirectional aggregate counts.
AI Subject Redundancy Filtering
Implementation of unique-subject logic to reduce double-counting. This involves tracking individual detections across frames to differentiate new arrivals from repeated observations of the same person or group.
Thank You
We appreciate Three Rivers Park District's willingness to participate in this pilot and engage with the findings. This study was designed in the spirit of open inquiry—testing whether vision-based monitoring can meaningfully contribute to the toolset available to land management professionals.