A deep dive into our literature survey, research gap, problem statement, objectives, methodology, and the technologies powering our solution.
Existing research into human-elephant conflict and smart deterrent systems informs our approach. Below are key findings from our literature review.
Prior work in IoT-enabled sensor networks has demonstrated the feasibility of remote wildlife detection. Geophone-based ground vibration sensing has been validated for detecting large mammals in forest environments, providing a non-invasive detection method suitable for edge deployment.
LoRaWAN (Long Range Wide Area Network) technology has emerged as a leading solution for IoT deployments in rural and remote areas with limited internet infrastructure. Its low-power, long-range characteristics (up to 15km in open terrain) make it ideal for conservation monitoring.
Studies using acoustic and seismic sensors combined with ML classifiers report high accuracy in distinguishing elephant movement signatures from other wildlife. TinyML models compressed with TensorFlow Lite achieve >85% accuracy on edge microcontrollers.
Research from India and Africa shows that acoustic deterrents — particularly bee sounds and ultrasonic pulses — are effective in redirecting elephant movement. Behavioral science supports a multi-modal deterrence approach combining sound and light stimuli.
Sri Lanka records among the highest HEC rates in Asia, with significant annual crop losses and casualties reported by the Department of Wildlife Conservation. Existing mitigation strategies (electric fences, watchers) are costly and unsustainable for rural communities.
Despite significant research in individual detection methods and wireless communication protocols, a critical gap exists in integrated, low-cost systems that combine:
Existing solutions are either too expensive for rural Sri Lankan communities, reliant on constant internet connectivity, or lack the automation needed for timely deterrence. Our research bridges this gap by developing an integrated, field-deployable system.
How can a low-cost, scalable IoT system using edge AI and LoRaWAN communication effectively detect elephant intrusions and trigger non-lethal deterrents in real-time within rural Sri Lankan communities that lack reliable internet connectivity?
Human-elephant conflict (HEC) is one of the most serious challenges in Sri Lanka. The current approaches to managing HEC — including electric fences, paid watchers, and ad hoc deterrents — are unsustainable, unreliable, and inaccessible to most affected communities.
This research addresses the problem of how to reliably detect elephant intrusions and automate deterrent responses in power-constrained, connectivity-limited rural environments, while keeping the system cost-effective enough for widespread adoption.
The research is guided by the following specific objectives:
Integrate geophone (seismic), PIR (passive infrared), and acoustic sensors to achieve reliable, non-invasive elephant detection at field boundaries.
Develop and deploy TinyML models on ESP32/Raspberry Pi that classify elephant presence from sensor data with high accuracy and low latency.
Build a robust long-range wireless communication backbone for transmitting intrusion alerts to farm owners and community members without internet dependency.
Integrate and automate sound (bee sounds, ultrasonic) and light-based deterrents that activate upon confirmed detection events to safely repel elephant intrusions.
Conduct controlled tests at Dehiwala Zoo and Pinnawala Elephant Orphanage, then validate in actual rural deployment environments in Sri Lanka.
Our methodology follows a layered approach — from sensing at the field edge, through AI-based classification, to communication and alert delivery.
Geophone and PIR sensors are deployed along farm perimeters. The ESP32 microcontroller handles real-time data acquisition and initial filtering, operating in low-power sleep mode between detection events.
Raw sensor signals are processed on the Raspberry Pi. A TensorFlow Lite ML model classifies incoming data as elephant, other wildlife, or environmental noise, minimizing false positives.
Confirmed detections trigger LoRaWAN packets transmitted to a central gateway. Fallback to Wi-Fi mesh or 4G LTE ensures message delivery even in degraded network conditions.
Upon confirmed detection, automated deterrents activate — bee sound emitters, ultrasonic speakers, and strobe lights work in combination to safely drive elephants back without harm.
All detection events are logged to a cloud backend for long-term analysis, model retraining, and community reporting dashboards accessible via mobile app.
ESP32 microcontroller, Raspberry Pi 4, Geophone sensors, PIR sensors, thermal cameras, LoRa modules (SX1276), bee-sound speaker arrays, strobe light units.
TensorFlow Lite for edge inference, TinyML for microcontroller deployment, Python-based model training pipeline, custom CNN and LSTM architectures for time-series sensor data.
LoRaWAN (primary), Wi-Fi mesh (secondary), 4G LTE (emergency fallback). TTN (The Things Network) for LoRaWAN gateway management and MQTT-based message brokering.
Cloud-based data logging for event history and analytics. RESTful API for mobile app integration. Automated retraining pipeline for continuous model improvement using field data.