Related Experiment Video
Updated: Jul 3, 2025
07:03
Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration
Published on: February 23, 2017
7.7K
Small Target Detection in Refractive Panorama Surveillance Based on Improved YOLOv8
Xinli Zheng1, Jianxin Zou1, Shuai Du1
1College of Science, Donghua University, Shanghai 201620, China.
Sensors (Basel, Switzerland)
|February 10, 2024
Summary
This study introduces an improved YOLOv8 model for detecting small targets in panoramic images from UAVs. The enhanced model boosts accuracy and efficiency for high-altitude surveillance applications.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Aerospace Engineering
Background:
- Panoramic imaging is crucial for Unmanned Aerial Vehicle (UAV) and high-altitude surveillance.
- Detecting small targets in wide-area, high-resolution panoramic images presents accuracy and real-time performance challenges.
Purpose of the Study:
- To propose an improved lightweight network model based on YOLOv8 for enhanced small target detection in panoramic images.
- To address limitations in accuracy and real-time performance for surveillance applications.
Main Methods:
- Developed an improved lightweight YOLOv8 network model.
- Implemented a panorama copy-paste data augmentation technique.
- Utilized an unfolding, cutting, and stitching process for panoramic image analysis.
Main Results:
- Reduced model size by 10.6% and parameter count by 11.69% while maintaining detection speed.
- Achieved a 2.9% increase in overall mAP@0.5 and a 20% improvement in small target detection accuracy.
- Data augmentation and image processing techniques further boosted mAP@0.5 by 0.6% and 21.3% for small targets, and 4.2% overall, respectively.
Conclusions:
- The improved lightweight YOLOv8 model effectively enhances small target detection accuracy in complex panoramic scenarios.
- The proposed data augmentation and image processing methods significantly improve performance for UAV and high-altitude surveillance.
