Detecting the drone by the combination of optical flow with various techniques

In image processing, detecting and tracking air targets (especially drones) is attainable. The present study focuses on drone detection using the combination of optical flow and histogram and presents a new approach to the recognition of flying targets (drones). Drones are mostly occupying a small portion of the visual field and are often moving in front of complex backgrounds such as buildings, crowded scenes, and local motion of the trees and leaves. In our work, video images obtained from YOUTUBE were used for considering these motions. Videos were transformed into N-frame by free video-tojpg converter software. Then, the drone boundary was determined with MATLAB software in the images containing the drone. Eventually, these pictures and drone-free pictures were applied as a database. In this work, four approaches were introduced: 1) combining the histogram of oriented gradient (HOG) and support vector machine (SVM), 2) combining the optical flow (OPF) with segmentation, histogram of oriented gradient (HOG), and support vector machine (SVM) without background elimination, 3) combining the optical flow (OPF) method, image segmentation, histogram of oriented gradient (HOG), support vector machine (SVM), and background elimination by K-means procedure, 4) combining the optical flow (OPF) method, image segmentation, histogram of oriented gradient (HOG), support vector machine (SVM), convolutional neural network (CNN), and background elimination by K-means procedure. Simulation outcomes demonstrated that the proposed procedures have acceptable performance in detecting one or several drones on the scene.