Vehicle identification and classification are essential for Intelligent Transportation Systems, although current approaches may face challenges due to data collection discrepancies. This research introduces a novel approach for analyzing sequences of aerial images to detect and classify vehicles. Our methodology comprises six essential stages: Initially, we focus on images improvement to mitigate noise and improve contrast. Secondly, we segment images to emphasize the essential foreground items, subsequently detecting vehicles using the EfficientNet approach. Following detection, we extract features employing several descriptors: FREAK, KAZE, and SURF. The KFDA optimizes these characteristics, after which the data will be categorized using a Neuro-Fuzzy Classifier. Our trials on the AU-AIR dataset achieved a mean accuracy of 91.49%, demonstrating that our approach surpasses conventional techniques. These results illustrate its use in monitoring aerial vehicles in practical scenarios and indicate significant potential for enhancing efficiency in traffic zone management and surveillance systems.
Drone-Based Road Traffic Surveillance: Multi-Vehicle Tracking and Classification
2024-11-20
927910 byte
Conference paper
Electronic Resource
English
Automatic Traffic Surveillance System for Vehicle Tracking and Classification
Online Contents | 2006
|Computer Vision-Based Intelligent Traffic Surveillance: Multi-Vehicle Tracking and Detection
BASE | 2024
|