Research On Tracking Algorithms For Infrared Image Terminal Guidance Based On Moving Target

With the urgent needs of our precision strike weapons, to carry out the researchon tracking algorithms for infrared image terminal guidance based on moving targethas very important defense significance. At terminal guidance tracking stage, with thecamera gradually approaching to the target, the projected area of the target in theimage, the appearance of the target and background will change drastically.Commonly target tracking algorithm is difficult to meet the requirements, so, terminalguidance tracking is still a challenging research topic.In this paper, terminal guidance tracking is broken down into three phases,namely the dim and small target acquisition, the surface target tracking, and precisionstrike parts identification. Composite-tracking mode is adopted in this paper, mainlyincluding the following three aspects:(1) Dim and small target acquisition phase. The distance from target to thecamera is very far. The projected area of the target in the image is small, appearing ablurry spot. Such target generally does not have the shape and texture features. Forthis type of target, the acquisition algorithm based on the layered backgroundcompensation of the region of interest is proposed. First, image regions of interest areextracted, and layered optical flow parameter is calculated for each region of interest;then, establish background perspective motion model, and compensate the relative movement of the background; finally, adaptive difference approach is adopted todetect moving target. For further confirmation of the target, trajectory correlationalgorithm which makes use of the timing characteristics of the target motion is used,so that the probability of false alarm is greatly reduced.(2) Surface target tracking stage, the target and the camera distance graduallyapproaching. At this time the target is gradually showing detail and texture features.Because of the dramatic relative motion, the appearance of the target in the imagesequence changes drastically. So, using the fixed template to track is bound to leadingtracking failure. To solve this problem, the subspace tracking algorithm based onparticle filter is proposed. Which use sparse image representation and subspacefeature extraction, makes the system adaptive to changes in appearance of the target.First, particle filter is applied to importance sampling, and the sample set arerepresented by sparse image; then, project the sample set to target subspace, andestimate the maximum likelihood probability to update the target state; finally, updatethe target subspace. Thanks to its online learning and adaptation, that can make thesystem adaptive to the target change.(3) Precision strike parts identification phase. The target image in the field ofview becomes large, even full the field of view. In order to keep tracking processcontinues, the target recognition algorithm based on local robust features is proposed.That identifies the target partial crucial position as a new tracking point. Thetraditional feature point matching method exists failure matching. So, two-pointimprovement strategies are applied. The experiments indicate that the improvedalgorithm achieves good results, and does not affect the correctness of the premise.