IoT Based Elephant Monitoring and Early Warning System

Abstract

Human–elephant conflict (HEC) poses severe socioeconomic and safety crises globally, particularly in agricultural forest-edge regions. Traditional mitigation strategies, such as physical fencing and manual patrolling, are high-cost, labour-intensive, and prone to animal habituation. This study presents a decentralized, low-cost, multi-modal Internet of Things (IoT) early warning system powered by an off-grid solar direct current (DC) module. The framework utilizes a dual-stage, event-driven validation loop to maximize detection reliability while optimizing energy resources. Initially, a subsurface SM24 geophone sensor continuously samples micro-seismic vibrations caused by terrestrial locomotion. Ground noise is filtered locally using a software amplitude threshold of ±2,500 raw analog-to-digital converter (ADC) differential units, keeping power-intensive visual components in standby mode. Upon a verified threshold breach (t≥5s), a digital CCD camera module captures low-noise visual frames. Localized edge-inference is executed on an embedded Raspberry Pi using an optimized YOLOv8 deep learning network, achieving a validation accuracy of 88.1% mAP_50, a precision profile of 90.4%, and an out-of-sample test accuracy of 87.0%. Validated detections (≥70% confidence) trigger on-site visual strobes and directional high-decibel sirens playing biomimetic bee acoustics in under 7.0s. Simultaneously, real-time Markdown text alerts and visual threat payloads are dispatched via the Telegram Cloud Bot API alongside automated cloud tracking to Google Drive and Google Sheets. This prototype establishes a scalable, non-invasive intervention that secures local livelihoods and supports human–wildlife coexistence.