Detection, Tracking and Classification of Aircraft and Drones in Digital Towers Using Machine Learning on Motion Patterns

Abstract

Remote digital towers using high-resolution cameras that cover a 360-degree view of airports have recently been a solution to some airports in replacing physical towers and in introducing digital technologies as tools to assist tower controllers. We propose the use of machine learning techniques to automatically detect and track small moving objects in the airfield from their motion patterns, i.e. the ways an object moves. A video dataset comprising aircraft in an airfield and drones captured from a fixed angle is constructed. To the dataset, we apply Harris Detection and Convolutional Neural Network followed by Optical Flow to locate and track very small moving objects in the wide-area scene. Motion-based features are extracted from their trajectories after which a K-Nearest Neighbor classifier is applied to classify objects into drones or aircraft. Our tests show that the accuracy and execution time are appropriate for real-time operation.