An Automatic Traffic Control System over Aerial Dataset via U-Net and CNN Model

Vehicle detection is integral to advanced surveillance and traffic monitoring systems, which traditionally rely on cameras positioned on bridges or roadsides. Aerial imagery, however, offers enhanced versatility through mobile platforms capable of surveying expansive areas efficiently. Despite progress in existing models for vehicle recognition and tracking, challenges persist in maintaining high accuracy within complex road environments. This study introduces an innovative six-stage approach for vehicle detection and tracking in aerial image sequences. The process begins with georeferencing and image preprocessing to reduce noise and enhance brightness for optimal analysis. U-Net Segmentation is applied for image segmentation, followed by precise vehicle detection and localization using the powerful Mask-RCNN algorithm. Subsequent stages involve vehicle matching facilitated by ORB (Oriented FAST and Rotated BRIEF) feature extraction and tracking through the optical flow algorithm for robust motion continuity. The proposed method achieved a remarkable detection accuracy of 94 percent and a tracking accuracy of 89 percent on the AU-AIR dataset. Comparative evaluations on publicly available datasets demonstrate that this approach surpasses state-of-the-art methods, showcasing its potential applications in traffic monitoring, enhanced surveillance, and search-and-rescue missions.