Related Experiment Video
Updated: Jun 14, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
492
Vehicle recognition pipeline via DeepSort on aerial image datasets
Muhammad Hanzla1, Muhammad Ovais Yusuf1, Naif Al Mudawi2
1Faculty of Computing and AI, Air University, Islamabad, Pakistan.
Frontiers in Neurorobotics
|September 2, 2024
Summary
This study introduces an automated system for intelligent traffic monitoring using Unmanned Aerial Vehicles (UAVs). The method enhances vehicle detection and tracking accuracy in complex aerial scenes.
Area of Science:
- Computer Vision
- Artificial Intelligence
Background:
- Unmanned Aerial Vehicles (UAVs) offer efficiency in traffic monitoring but struggle with automated vehicle extraction from complex scenes.
- Automating vehicle identification and tracking in aerial imagery remains a significant challenge.
Purpose of the Study:
- To develop an advanced autonomous vehicle surveillance system for intelligent traffic monitoring.
- To improve the accuracy and efficiency of vehicle detection and tracking in aerial images.
Main Methods:
- Utilizes Fuzzy C-Means (FCM) for image segmentation and YOLOv8 for precise vehicle detection.
- Employs ORB features for robust vehicle recognition and assignment across frames.
- Integrates DeepSORT with Kalman filtering and deep learning for accurate vehicle tracking.
Main Results:
- Achieved a precision of 0.86 for vehicle detection on the VEDAI dataset and 0.84 on the SRTID dataset.
- Demonstrated vehicle tracking accuracies of 0.89 on VEDAI and 0.85 on SRTID.
Conclusions:
- The proposed method significantly enhances autonomous vehicle surveillance capabilities.
- The system offers a robust solution for intelligent traffic monitoring using UAVs.
