Classification of Features Extracted from Image Foreground and Background for Tracking of Aerial Moving Targets

Moving target tracking is a process in which an object is tracked and its location is determine in each frame. The goal of this process is facilitating the subsequent processing to analyze the behavior or detect moving objects. In this paper a new approach for aerial moving targets tracking based on feature matching algorithms have been proposed. The main challenge is classification of extracted features from background and foreground regions. To solve this problem, key points and their corresponding on the patterns extracted from consecutive frames, is calculated by the KLT algorithm. Then for each of this points, six attributes such as color, mean, variance and range of variation are calculated. Using these attributes and Bayesian discriminant function, extracted key points classified. In addition to resisting the proposed algorithm to scale change of target the object history scale in the 10 previous frames is used. Proposed algorithm was performed on an AIRCRAFT TRACKING standard database. Experimental results show the effectiveness of the proposed method against KLT and SURF tracking algorithms in term of accuracy.