Detection of air traffic conflicts in a weather constrained airspace is challenging given the inherent uncertainties and aircraft maneuvers which give rise to new conflict birthpoints. Traditional conflict detection tools are untenable in such situations as they primarily rely on flight-plan, aircraft performance characteristics and trajectories projection in shortterm (2-4 minutes). This work adopts a convolutional neural network (CNN) model, on radar-like images, for conflict detection task in a constrained airspace. The CNN models are well-known for their learning capabilities when dealing with unstructured data like pixelated images. In this study, historical ADS-B data with weather constrained airspace is input as pixelated images to the CNN model. The learned model was compared with two well-known models for conflict detection (CD). The results demonstrated that the CNN based model was able to predict off-nominal conflict with high accuracy. The CNN model also demonstrated its ability to predict off-nominal conflict early for a given ten-minute look-ahead window. The CNN based model also showed low levels of false alarm signals as compared to other models. Generally speaking, all models showed low probabilities of miss-detection, mostly in the early phase of the 10-minute look-ahead window. This novel approach may serve to develop effective CD algorithms with longer look-ahead time and may aid in early detection of air traffic conflicts in non-nominal scenarios.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.