Averting Human-Elephant Conflict using IoT and Machine Learning of Elephant Vocalizations

Abstract

Human-Elephant conflict causes harm to life and property and has contributed to endangering elephants. Elephant deterrent systems have been built using electric wire fences [1], improving the fences with chili [2], and sound deterrents to scare the elephants [3] and all have limited success [4]. The elephants are hunted and killed making them endangered [5]. The elephants help the ecosystem [6]. This paper outlines the approach we took toward a more sustainable world where humans can co-exist with elephants by building an elephant alertness system to predict when they are dangerously close and to alert humans. We built an IoT device using bio-acoustics and machine-learning as an early warning system to determine the proximity and behavior of elephants by classifying elephant vocalizations. We built this early warning device using a Raspberry Pi along with a microphone and an alarm system. The device identifies the presence of elephants nearby using sound sensors. We built an AI machine learning model that identifies the type of vocalization as a Chirp, Roar, Rumble, or Trumpet and predicts whether the elephants are likely to raid even when they are not visible from darkness or thick foliage. In this paper, we share the challenges we solved in building this outdoor weather-friendly Artificial Intelligence IoT (AIoT) device hosting the artificial intelligence predictive model.