Study On Target Tracking Method Based On Multi-source Image Fusion In Complex Backround

Target tracking under complicated environment based on multi-sourceimage is very popular nowadays. Illumination change, target change theshape itself and stealth or covered by other stuff caused by the backgroundwill all lead to error of target tracking. Visible image can express detail ofthe target and is sensitive to a lot of information like color. It has greatdescription on shape and texture but when the target is stealth or coveredby other stuff, it will lose information. Infrared image is used to describethermal radiation. It could express the whole target clearly and have betterpenetration than visible image but it will lose details. Therefore, visibleand infrared image need to be fused and play to their strengths.Using the tracking analog measurement system to analog targetmoving in complicated environment and get the visible and infrared imagesequences to do research on the fusion tracking algorithm based on the twokinds of image sequences.First, image preprocess need to be done. By analyzing the result ofimage smooth, image segmentation and morphological filter algorithm,using median filter combined with OTSU adaptive threshold segmentationalgorithm and open operator could separate the target from the backgroundeffectively. Then background difference method is used to get the detectlocation of the target.Based on target tracking algorithm based on sparse representation, wechoose a tracking algorithm based on sparse measurement matrix andBayesian classifier of two classes. It use sparse measurement matrix to do dimensionality reduction. It could not only reserve the basic information ofthe target but reduce trial information. Thus, it will not only guarantee theaccuracy of the result, but also greatly reduces the complexity of thealgorithm and improve the computing speed. By deciding the result of theBayesian classifier, the sample with most similarity is chosen to be thetracking result and be used to update the classifier. Despite the perfectresult for some large target with more information of this algorithm, itcould not perform well on the surveillance video OTCBVS database.Visible and infrared image should be fused to get better result.Fusion tracking algorithm could be divided into three levels, pixellevel, feature level and decision level. By using the tracking algorithmbased on sparse measurement and Bayesian classifier, an adaptive decisionlevel fusion tracking algorithm is proposed. The tracking result of the twodifferent image sequences are used to do the decision and choose the bestresult then do the classifier update. This algorithm could solve the trackingfail problem.Based on the tracking analog measurement system in lab, a softwaresimulation platform is developed to achieve the method in this paper. It canshow the preprocess algorithm, target detection and tracking algorithm ofsingle sensor and multiple sensors. It will build a solid foundation forfurther researches.