Performance Analysis and Evaluation of Object Detection Algorithms for Drone Networks

The YOLOv5 and YOLOv8 object identification algorithms are optimised in this thesis for edge devices such as the Raspberry Pi 4, which are becoming more and more important in decentralised computing environments where real-time processing capabilities are essential. The main goals are to guarantee high accuracy and low latency in applications like active surveillance and driverless driving, while also improving the operational efficiency of these algorithms on resource-constrained edge devices. Reducing power consumption is essential to this optimisation since it increases the lifespan of devices in distant or mobile environments where energy supplies are limited. Adapting these algorithms to edge-specific frameworks such as OpenVINO and NCNN is the goal of the research, which tries to preserve object detection integrity without sacrificing speed or power economy. The thesis assesses the performance trade-offs associated with these modifications by methodical testing, offering insights into how speed, precision, and power economy are balanced. The results could establish new standards for the efficient deployment of intelligent systems in resource-constrained environments, making a substantial contribution to the domains of autonomous technologies and real-time data processing.