Drone Detection and Classification using Computer Vision

Abstract

Drones, also known as Unmanned Aerial Vehicles (UAVs) are based on the principle of rotor torque pushing the air down which results in the upward lift of the drone. UAVs are used in tasks such as rescue operations and item delivery in remote areas, surveillance, agriculture, wildlife conservation, outer space and photography. Due to their low cost and high efficiency, it is used by diverse groups for both better and worse causes. The cases of malicious uses of military drones and spy drones employed are on a rise. The malicious activities deployed by the military drones may include air strikes at enemy military bases, army troops and in some cases end up causing the death of civilians in proximity. Drones that are used for espionage can retrieve valuable information regarding the different strategies of the military, can track the location of the army personnel and spy upon unsuspecting civilians. These drones are eliminated after being detected by the persons involved but sometimes in order to reduce risks, harmless delivery drones are also discarded. In order to aid against the malicious drone activity while making sure that unnecessary panic over the delivery drones and material-loss is not caused, a real-time computer vision system is proposed that can identify the drone in the given region of interest, give its relative location and classify the drone. The Convolutional Neural Network (CNN) architecture, You Only Look Once Algorithm (YOLOV5), is used to classify the drone into one of the categories: Army, Surveillance and Delivery drones.