A New Scheme for Vision Based Flying Vehicle Detection Using Motion Flow Vectors Classification

This paper presents a vision based scheme for detecting flying vehicle using a new feature extraction and correspondence algorithm as well as a motion flow vectors classifier. The base of detection is to classify the motion flow vectors of object and scene at two video sequences from a mobile monocular CCD camera. For this purpose, we introduce a method to extract robust features from fuzzified edges at first frame. Then, correspondence features are approximated at second video frame by a multi resolution feature matching processing based on edge Gaussian pyramids. In next stage, the estimated motion flow vectors classify into two object and scene classes using a supervised machine learning method based on MLPs neural network. In final step, the flying vehicle localize by approximating the contour of object based on a convex hull algorithm. Experimental results demonstrate that the proposed method has proper stability and reliability especially for the detection of aerial vehicle in applications with mobile camera.