End to End Real Time Drone Based Person Detection Framework Using Deep Learning

Jul 16, 2026

Upload by Riley

Rapid-response UAV operations for security, search and rescue, and surveillance demand consistent person detection under large scale changes caused by altitude and viewpoint. The work proposes an integrated real-time detection pipeline that performs target detection from a wireless live drone video feed. It builds on the YOLOv8-nano architecture and uses flight experiments across multiple altitudes. Trained on VisDrone2019, the model reaches 57.4% precision, 41% recall, 44.8% mAP, and 20.3% mAP50:95, while real-world tests show near-total reliability at 16–25 m with stable speed above 41 FPS and a peak of 50 FPS.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

Introduction

Motivation and Applications

Related Work and Limitations

Proposed Real-Time UAV Detection Pipeline

End-to-End Real-Time Drone-Based Person Detection Framework Using Deep Learning 1st Payel Sarmah Centre for Drone Technology Indian Institute of Technology Guwahati Guwahati, India [p.sarmah@iitg.ac.in](p.sarmah@iitg.ac.in) 2nd Ayush Ranjan Dept. of CSEKIIT University Bhubaneswar, India [ranjanayush881@gmail.com](ranjanayush881@gmail.com) 3rd Piyush Kaushik Bhattacharyya Dept. of CSEKIIT University Bhubaneswar, India [piyushbhattacharyya@gmail.com](piyushbhattacharyya@gmail.com) 4th Anil Kr. Shaw Drone LAB National Institute of Electronics and Information Technology Bhubaneswar, India [anilshaw2785@gmail.com](anilshaw2785@gmail.com) 5th Pradip Kr. Das Centre for Drone Technology Indian Institute of Technology Guwahati Guwahati, India [pkdas@iitg.ac.in](pkdas@iitg.ac.in) arXiv :2607 . 10605v1 [ cs .CV] 12 Jul 2026 Abstract—In recent years, Unmanned Aerial Vehicles (UAVs) or drones have gained rapid response in terms of security, search and rescue (SAR), border surveillance, etc. Existing monitoring frameworks often struggle to maintain detection consistency when targets undergo significant scale variations due to altitude changes, leading to critical information gaps. To address this issue, this work proposes an integrated realtime detection pipeline for detecting targets through the wireless live drone video feed. Build upon YOLOv8-nano architecture, extensive flight experiments were conducted to determine the detection performance across multiple flight altitudes. Trained on VisDrone2019 dataset, the results of YOLOv8-nano model achieves 57.4%, 41%, 44.8% and 20.3% in precision, recall, mAP and mAP50:95 respectively. While demonstrating on real environment, this analysis revealed that the algorithm achieves near-total detection reliability at altitudes between 16 and 25 meters with the detection frame rate consistently maintained above 41 FPS and reaching a peak of 50 FPS. However, the goal of this work is to enable real-time person detection from an aerial platform via wireless transmission. This approach effectively addresses the dual challenges of identifying targets at varying scales and ensuring near-to-accurate localization during aerial observation. Index Terms—Person Detection, YOLOv8, Deep Learning, Real-Time, UAV. I. INTRODUCTION Real-time person detection from aerial platforms has become an important component of modern intelligent surveillance systems. It is increasingly used in applications such as disaster management, building monitoring, border security, large-scale crowd monitoring, etc. where rapid situational awareness is essential. Compared to ground-based sensors, Unmanned Aerial Vehicles (UAVs) offer a broader field of view and flexible deployment, allowing operators to observe complex scenes that would otherwise be difficult or unsafe to access. In emergency situations, this capability enables first responders to evaluate affected regions from a safe distance, while in public security scenarios, aerial monitoring supports efficient oversight without direct physical presence. Traditional approaches to person detection relied heavily on manual observation or classical computer vision techniques. Methods such as background subtraction [1], frame differencing [2] and Gaussian mixture models [3] were effective in controlled settings with static cameras. Yet these techniques degrade rapidly when deployed on UAVs. The continuous motion of the drone introduces global scene changes that render background subtraction ineffective. Furthermore, aerial imagery presents unique difficulties including varying altitudes, steep viewing angles and rapid illumination changes caused by cloud cover or shadows. These factors often lead to missed detections when using non-learning-based algorithms. The paradigm shift toward deep learning has successfully addressed many of these limitations. Convolutional Neural Networks (CNNs) have become the standard for visual perception tasks. Specifically, the YOLO (

FAQs

What problem does the proposed UAV person detection framework address?

It targets inconsistent detection performance caused by significant scale variations from altitude and viewpoint changes during aerial observation, which can create critical information gaps.

Which model architecture is used for detection and how is it evaluated?

The framework builds on the YOLOv8-nano architecture and is evaluated through extensive flight experiments across multiple flight altitudes, using a wireless live drone video feed.

What detection performance and operating altitude range are reported?

Trained on VisDrone2019, YOLOv8-nano achieves 57.4% precision, 41% recall, 44.8% mAP, and 20.3% mAP50:95. In real environments, near-total detection reliability is reported at altitudes between 16 and 25 meters with frame rates maintained above 41 FPS and peaking at 50 FPS.

Jul 16, 2026

Upload by Riley