Enhancing Airside Monitoring: Multi-view Approach for Accurate Aircraft Distance-To-Touchdown Estimation in Digital Towers
Primary Area: representation learning for computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Runway Management; Multiple Cameras; Airside Monitoring;Distance-To-Touchdown Prediction
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: A real-time deep learning method is introduced to estimate the distance of approaching aircraft, improving airport and runway operations using digital towers and advanced computer vision.
Abstract: A digital tower, a cost-effective alternative to physical air traffic control towers, is expected to provide video-sensor-based surveillance, which is particularly advantageous for small airports. To fully realize this potential, advanced computer vision algorithms play a crucial role in effective airside monitoring. While current research primarily focuses on tracking aircraft on airport surfaces, an equally vital aspect is the real-time observation of approaching aircraft on the runway. This capability holds a pivotal role in augmenting both airport and runway operations. In this context, the study introduces a real-time deep learning approach to accurately estimate the distance-to-touchdown of approaching aircraft, covering distances of up to 10 nautical miles. The approach overcomes the limitations of monoscopic and stereoscopic methods by utilizing multi-view video feeds from digital towers. It integrates Yolov7, an advanced real-time object detection model, with auxiliary regression auto-calibration, enabling real-time tracking and feature extraction from diverse camera viewpoints. Subsequently, an ensemble approach utilizing an LSTM model is proposed to combine input vectors, resulting in precise distance estimation. Notably, this approach is designed for easy adaptation to various camera system configurations within digital towers. The model's effectiveness is assessed using simulated and real video data from Singapore Changi Airport, demonstrating stability across scenarios with low predictive errors (Mean Absolute Percentage Error = 0.18%) up to 10 nautical miles under visual meteorological conditions. These capabilities within a digital tower environment can significantly enhance the controller's ability to manage runway sequencing and final approach spacing, ultimately leading to remarkable airport efficiency and safety improvements.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2476
Loading