Research On Defined Target Locking And Tracking In Aerial Images

Under the condition of aerial shot, target locking and tracking is the main processing method ofprecision strikes, and has the vital significance to our national defense development. In complex aerialimages, the background is moving between moving target and image background because of themotion of camera platform. So the target locking and tracking in aerial images has been getting morecomplex. The main contents of this dissertation are as follows:First, the research on background motion compensation is deeply discussed. For thecharacteristic of background movement in aerial image sequence, a new method based on theimproved MIC algorithm is proposed to estimate the motion vectors of the image feature pointsquickly and find the matching points exactly. Combining the robust RANSAC algorithm, the affinetransform parameters of the background motion can be estimated, and then the background motioncan be compensated. Based on the background motion compensation, using an improved algorithm offrame difference method and combining morphological operation, moving target can be detected andlockeTdo e xenascutlrye. real-time target tracking of the aerial images, the mean shift algorithm is studied, andin the light of the complexity of the aerial images, traditional mean shift tracking is improved via astrategy to improve and update the target model so that the ability of locking and tracking specialtargets in changeable circumstance is then enhanced. The real-time ability and reliability of theimproved algorithm has been verified in experimental simulation.To tackle the occlusion problem in aerial images, the target tracking technology based on particlefiltering is studied and improved. The tracking performance of the particle filtering is improved viathe model of a kernel-histogram in HIS space, and the degeneracy problem is also resolved by usingthe resampling method. Then the particle filter algorithm is extended to multiple target tracking.Theexperimental results have shown that the improved algorithms can effectively deal with occlusionproblem, and accurately track any given target on the ground.Finally, the tracking performance of the two algorithms are compared and analyzed, and theprospect of how to combine with particle filtering and mean shifting for improving the practicabilityof locking and tracking are presented.