Abstract
Large dairy farms are growing fast due to increased consumer demand, but this expansion brings issues such as reduced animal welfare and sustainability. Precision Livestock Farming (PLF) is needed to make informed decisions on animal health, feeding, breeding, and environment management. This study is based on GALA, an automated IoT system for monitoring the barn, developed in partnership with the University of Milan. It introduces a real-time localization tag embedded in a collar with accelerometer to predict behavior and improve localization. It also designs an efficient multi-level real-time processing pipeline with attention to power consumption and practical deployment.
Key Contributions
- Developed an algorithm that fuses accelerometer data with location data to improve position accuracy inside the barn.
- Extended collar firmware so the device acts as both behavior sensor and localization tag.
- Designed a multi-level architecture for real-time data ingestion, behavior prediction, and localization with low latency and energy usage.
- Validated system: achieved ~78% accuracy for tag localization within a 1.5m range, and ~90% accuracy for zone assignment based on cow behavior.
Tech Stack & Architecture
Hardware:
- Collar-mounted device with accelerometer and wireless communication.
- Real-Time Location System infrastructure in barn.
Firmware & Embedded:
- Lightweight firmware on collar for accelerometer sampling, firmware for tag functionality.
- Energy efficiency optimizations (duty cycling, low-power modes).
System Architecture:
- Real-time data pipeline: sensor → tag → processing unit with accelerometer fusion and localization.
- Middleware for ingesting the real-time data.
- Backend storage for behavior logs and localization zones.
Algorithms / Models:
- Sensor fusion algorithm combining location estimates + accelerometer-based behavior prediction.
- Zone classification model for behavior based on accelerometer patterns.
Results & Impact
- Localization accuracy: 78% within 1.5 meters; zone assignment accuracy: ≈90%.
- System suitable for real barn deployment: low power requirements, real-time behavior insights.
- Potential to improve animal welfare and farm management through better monitoring.
Status
This was a Master’s thesis (Politecnico di Milano, Academic Year 2022-2023), by Giuseppe Tortorelli, in collaboration with the Department of Agricultural and Environmental Sciences, University of Milan.