The remote digital tower offers advantages over traditional physical towers, while accurately locating objects in panoramic images presents challenges due to large variations in object scale and potential image quality issues. This paper proposes a simple yet effective cross-scale object detection method for panoramic images in the context of remote tower systems for air traffic control. Our approach focuses on adaptive feature adjustments to enhance small object representation and explores higher computational efficiency in convolutional layers for improved network inference. Additionally, we utilize TensorRT for inference, optimizing the network structure and employing GPU hardware acceleration. Results show that the proposed method can significantly enhance cross-scale object detection performance, contributing to the advancement of remote tower systems.
Cross-Scale Object Detection for Large-Scale Images in Real-Time
2023-10-11
2581048 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Real-Time Recognition of Large-Scale Driving Patterns
British Library Conference Proceedings | 2001
|Real-time recognition of large-scale driving patterns
IEEE | 2001
|Large-scale processing and querying for real-time surveillance
Europäisches Patentamt | 2017
|