A Study of Detection and Tracking of Artificial Intelligence in UAVS using Machine-Learning Approach - Machine Learning Approach Evaluation

Aug 3, 2026

Upload by Ivy

Research develops and tests an automated detection system for multi-rotor unmanned aerial vehicles (UAVs) to support protection of sensitive structures and privacy preservation. Initial trials with computer vision using the ORB feature detector showed limited real-world suitability, leading to a machine-learning detection strategy. Accuracy is improved by augmenting the Common Objects in Context dataset with 1000 UAV samples from the Safe Shore dataset. Four experiments evaluate single- and multi-UAV scenarios in static images and videos, achieving 97.3% detection success under optimal conditions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Introduction

UAV visual and motion characteristics

Method and Approach

ORB-based computer vision baseline

Machine-learning detection strategy

Dataset augmentation and training

Experiments and Evaluation

Static image scenarios

Video footage scenarios

Single-UAV vs multi-UAV testing

Results

Detection performance under optimal conditions

A Study of Detection and Tracking of Artificial Intelligence in UAVS using Machine-Learning Approach Akshada Pandurang Kulkarni Ph. D. Scholar Department of Computer Application Dr. A. P. J. Abdul Kalam University, Indore MP. [a](akshada.ma@gmail.com)[kshada.ma@gmail.com](akshada.ma@gmail.com) Dr. Atul Dattatreya Newase Research Supervisor Department of Computer Application Dr. A. P. J. Abdul Kalam University, Indore MP. [dr.atulnewase@gmail.com](dr.atulnewase@gmail.com) Abstract—This article explores the development and testing of a system for detecting multi-rotor unmanned aerial vehicles (UAVs), a critical need in sectors focused on safeguarding sensitive structures and preserving privacy. Initially employing computer vision techniques, specifically the Oriented FAST and Rotated BRIEF (ORB) feature detector, the study found its real-world applicability limited, prompting a shift towards a machine-learning based detection method. To enhance the model's accuracy, the Common Objects in Context dataset was supplemented with 1000 UAV samples from the Safe Shore dataset. The system's efficacy was evaluated through four rigorous experiments, encompassing scenarios with a single UAV and multiple drones captured in both static images and video footage against sky backdrops. Achieving a notable detection success rate of 97.3% under optimal conditions, this study demonstrates the potential of integrating advanced machine learning techniques with enriched datasets for reliable UAV detection in diverse operational environments. Keywords-UAV; detection; machine learning; Tensor Flow; ORB. I. INTRODUCTION Algorithms for object identification and classification rely on the fact that the items under consideration share certain features. So characteristics aren't only what they seem like; their activity or movement also defines them. UAV recognition is concerned with both things that may be found inside the sensing region and with UAVs themselves. A UAV's form is one of its most distinguishing visual characteristics. From the tricopter to the octocopter, every kind of unmanned aerial vehicle (UAV) looks the same. tricopters are triangular in form, quadcopters are square, and so forth. Additionally, each UAV is made up of a unique set of hard components that give it its own distinct aesthetic appearance. Depending on the number of propellers, this design may have anywhere from three to eight arms with storage space in the centre for the control board. The propeller is hung from the engine, which completes each arm. Almost every UAV may utilise this basic appearance design since it is universal. UAVs in the nano and micro size categories, however, may have their propeller and motors placed directly on the UAV's centre panel. A major benefit of these machines is their longer battery life, which allows them to stay in the air for longer periods of time while also flying greater altitudes. Multi-rotor and fixed-wing drones can both take off and land from the ground, but most can only remain in the air for as long as they're moving. Another kind of drone exists that is smaller, lighter, and has a higher top speed than multirotor versions. In the drone industry, they are referred as as fixed-wing drones [3] . This is because of their characteristics, which make them popular in fields including environment and area mapping (with the potential for three-dimensional data production [4]), meteorology, and quality control inspections (an interesting example is the inspection of electrical wiring described in [5]) . According to [1,] fixed-winged UAVs are well-suited for the aforementioned tasks due to their low cost of ownership, low operating costs, and short flight duration. Their operational range and flying safety are also major advantages, according to [2] . One benefit of multi-rotor models is that they allow for more precise environmental mapping, which is useful in civilian applications. Another advantage is tha

FAQs

Why was ORB-based computer vision replaced in the study?

ORB feature detection showed limited applicability in real-world conditions, so the work shifted to a machine-learning based detection method.

How was the training dataset enhanced for UAV detection?

The Common Objects in Context dataset was supplemented with 1000 UAV samples from the Safe Shore dataset to improve model accuracy.

What kinds of experimental scenarios were used to evaluate detection performance?

Evaluation included scenarios with a single UAV and multiple drones, tested on both static images and video footage against sky backdrops.

Aug 3, 2026

Upload by Ivy