KLT BASED PREDICIVE AIRCRAFT IN LOW RESOLUTION IMAGES

Abstract

Tracking of a particular object in an image using the feature tracking algorithms faces significant challenges under conditions of severe aircraft rotation, highly cluttered background presence, arrival and collision with other aircrafts and sun, excessive noise, and varying lightening conditions due to weather changes. In this research, a real-time and robust algorithm is presented for the tracking of an aircraft in video sequences of low resolution with the use of Kanade-LucasTomasi (KLT) feature tracker, random sample consensus (RANSAC) algorithm, texture modeling and matching techniques. Features of an aircraft are tracked using KLT feature tracker while developing aircraft model at each frame and finding transition matrix with the use of each pair of consecutive frames. Features of aircrafts are excluded when not following the transition of the aircraft and new features are introduced if they are within the bounding range of the aircraft and match with the current aircraft model. The varying weather conditions and noise reduction are handled in the algorithm without the use of averaging filters to increase the overall efficiency of the framework.