A NEW APPROACH USING COMPUTER VISION FOR DRONE DETECTION

Abstract

Nowadays, one individual or organization can easily get a drone with an affordable budget. With the ability of carrying explosive materials, cameras and illegal things, drones can become security threats to military and civilian organizations. The detection of drones appearing in unauthorized areas becomes an urgent problem. This paper conducts empirical studies on training the deep convolutional neural network Faster R-CNN so that Faster R-CNN after training can most accurately detect drones in images. The obtained Faster R-CNN after training can then be used in drone detection, warning and defense systems for sensitive areas. Faster R-CNN is trained using a dataset of images with drone labeled bounding boxes and different training options. With proper training options determined through experiments, Faster R-CNN after training can detect drones with the average precision up to 0.774, which is 83% higher than Fast R-CNN with the average precision of 0.420 on the same dataset.