The rapid pace of urbanization and the growing complexities of modern traffic networks necessitate innovative and adaptive solutions to tackle critical challenges such congestion, safety, and sustainability. Integrating advanced technologies like aerial imagery and deep learning essential for optimizing traffic management as cities toward smart infrastructure. This study introduces a cutting-edge aerial-based vehicle detection and tracking framework, utilizing the AU-AIR and Aerial Car datasets to exceptional performance. The framework begins preprocessing, employing adaptive histogram equalization to enhance image quality and retain key vehicle features. Segmentation is performed using PSPNet, leveraging its multi-scale feature representation to accurately delineate boundaries. Vehicle detection is powered by providing unparalleled speed and accuracy in dynamic, traffic conditions. Vehicle matching is facilitated by a robust optical flow-based algorithm, ensuring consistent tracking even under occlusions. Vehicle counting is achieved using a Transformer-based model, offering precise and scalable traffic estimation. Vehicle tracking is optimized with a Kalman filter, maintaining smooth and reliable trajectories across Finally, the trajectories of tracked vehicles are yielding valuable insights into movement patterns and traffic flow dynamics. The proposed system demonstrates remarkable performance, with 97% detection accuracy and 93% tracking accuracy, establishing a new benchmark for aerial traffic monitoring. This framework has profound implications for the future of smart cities, intelligent transportation systems, and sustainable urban mobility.
Drone Based Traffic Surveillance using Semantic Segmentation and DeepSort
30.12.2024
838942 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch