Published Online:September 2025
Product Name:The IUP Journal of Information Technology
Product Type:Article
Product Code:IJIT030925
DOI:10.71329/IUPJIT/2025.21.3.41-58
Author Name:P Anjani Kumar, Kodavalla Durga Avinash, Potturi G V S Asrith and Yerasi Venkata Nithin Reddy
Availability:YES
Subject/Domain:Engineering
Download Format:PDF
Pages:41-58
Animal intrusion poses serious threats to agriculture, transportation, and human settlements in remote villages and base camps, leading to crop damages, economic losses, road accidents, and human fatalities. Traditional methods of physical fencing and deterrents have proven inadequate in mitigating these challenges. Since animals learn to bypass the traditional restrictions, there is a need for effective strategies for human-wildlife conflict mitigation. This paper proposes a machine learning (ML)-based approach for real-time animal intrusion detection using the single shot multi box detector (SSD) architecture. Using a custom dataset of different animal classes and humans, the detection model has been developed and further optimized for deployment on ARM-based devices. The system processes live video feeds from webcams or USB cameras, enabling rapid identification of intruding animals with bounding boxes and classification labels. The experimental results show detection accuracy ranging from 65% to 85% across most classes, with speed suitable for low-power Internet of things ( IoT) devices. While SSDcompromises slightly on accuracy compared to heavier models such as convolutional neural networks (CNNs), its computational efficiency makes very suitable for applications where processing resources are limited. The proposed framework provides a cost-effective and scalable solution for mitigating human-wildlife conflict, besides reducing crop damage, preventing road accidents, and enhancing safety in remote villages.
Animal detection has evolved significantly over the years, integrating traditional ecological techniques with cutting-edge computational methods. Historically, researchers relied on manual observations, footprints, and body patterns to identify and monitor animals. These methods, while foundational, were time-consuming and required skilled personnel.