The increased usage of unmanned aerial devices for commercial usage, such as drones, has presented a new challenge in aircraft security and overall public safety. Therefore, there is an urgent need to accurately detect and track drones. The objective of this paper is to classify rotary drones and fixed wing drones based on their trajectories. In order to develop classification models, Stone Soup open-source software framework is used to generate simulated track data using the location information of the drones available through Global Position System (GPS) telemetry data. Stone Soup can be used to study the quality of the tracks when classifying drones. To study the performance of the various classification methods in a realistic environment, false alarms were generated along with the tracker outputs. Tracker output is segmented into sub-trajectories and were used as inputs to the different classification models. Traditional machine learning algorithms namely Support Vector Machines (SVM), K-Nearest Neighbour (KNN), Decision Trees (DT) and a deep learning algorithm namely Convolutional Neural Networks (CNN) were considered for developing classification models. Kinematic features derived from the sub-trajectories were used as features for machine learning algorithms while images obtained from the sub-trajectories were used as input to the CNN. In order to handle class imbalances, data augmentation was used. The performance of the various classification models validated the objective of this paper.
|