Deep learning and fuzzy logic-based hybrid framework for aerial object tracking

Abstract

This study introduces an aerial target detection and tracking system using the YOLOv8 model combined with the ByteTrack algorithm. The proposed system is built on improving accuracy and efficiency in the detection and tracking of aerial objects in video frames. Detection is performed using a specially trained YOLOv8 model. ByteTrack further completes the tracking in a robust manner, even for dynamically changing environments. The presented scheme also embeds a new decision-making layer by using a Fuzzy Logic System. This system adopts Trapezoidal Membership Functions for confidence and distance evaluation of detected objects. It enables the framework to achieve priority levels in tracking. For prediction of future positions, a Kalman Filter is applied. This enhances the ability of the system to foresee various situations. Different scenarios effectively demonstrate how the system can dynamically prioritize and track multiple objects in accordance with their threat level and proximity. The findings of this study could contribute to existing methods by improving detection and tracking accuracy. They also incorporate a decision-making process like humans. Hence, its highly applicable domains are defense and surveillance, which require real-time and accurate threat assessment. This system can operate reliably under different operational conditions and may provide a valid tool for enhancing airspace security