Drone versus Bird Flights: Classification by Trajectories Characterization

This paper presents an alternative approach to identify and classify the small flying drones from the birds in near field, by examining the pattern of their flight paths and trajectories. The trajectories of the drones and birds were extracted from multiple clips of videos including various natural and synthetic database. Small drones that are piloted both automatically and manually usually fly in a stable manner. Their flight paths are usually straight and smooth with sharp angle turns. Whereas birds being natural flyers have flight paths that are intrinsically periodic due to their flapping motion with occasional straight glides and soaring sections. Five (5) trajectories characteristics and observed from the object's flight paths: turning angle, periodicity (frequency), curvature, object pace (velocity and acceleration). Subsequently, principal component analyses were applied to reduce the number of these trajectory features from 5 to 2 parameters. Classification by support vector machine with Non-linear transformation kernel was used. Sample test results show that the prediction was ≥80% accurate even at confident level of 90%. These predictions can be improved by gather more training data to cover more cases.