AI-Based FPV Drone System for Autonomous Tracking of Hazardous Environmental and Energy Objects

Abstract

Monitoring hazardous environmental and energy infrastructure zones - such as gas pipelines, high-voltage power lines, and industrial facilities - requires continuous aerial surveillance and autonomous object tracking capabilities. Traditional approaches rely on manual drone piloting or cloudbased processing, which increases operational latency and limits platform autonomy in remote environments. This paper presents the architecture of an AI-based FPV drone system designed for autonomous monitoring and tracking of hazardous environmental and energy objects using an embedded vision pipeline. The proposed system integrates a Raspberry Pi 5 onboard processing module, a Pixhawk 6C PX4 autopilot, and a hybrid tracking algorithm combining correlation-based visual tracking with AI-assisted detection capabilities. The system supports real-time FPV video streaming to a smartphone interface, enabling the operator to designate a surveillance target directly from the live feed, after which the drone autonomously maintains the target within the camera field of view using velocity-based MAVLink control commands. The proposed architecture outlines the feasibility of deploying autonomous monitoring drones on low-cost embedded hardware without GPU acceleration or cloud infrastructure.