Region Based Convolutional Neural Network for Human-Elephant Conflict Management System

Human elephant conflict occurs due to migration of elephants from their habitat to human living areas in search of food and water. In order to reduce the Human-Elephant Conflict, a real time prototype is built to migrate the elephant to human living areas is minimized by generating honey bee sound and tiger growl sound to which the elephant’s dislikes. Four object detection algorithms such as SSD mobilenet v2 model, SSDlite mobilenet v2 model, SSD inception v2 model, and Fast R-CNN inception v2 are considered. SSDlite mobilenet v2 model produced the best results with precision = 0.854 AP, recall = 0.718 AR, f1-score = 0.780, prediction time = 34.49ms for a frame rate = 31.15fps. Real time implementation is carried out using Raspberry Pi 3 with SSDlite mobilenet v2 architecture.