Rasearch On Object Tracking Algorithm In Dynamic Image

Moving object tracking in dynamic image sequences fuse advanced technology inmany fields, including image processing, pattern recognition, artificial intelligence,automatic control, et al, which has been applied very widely in the fields of industrialcontrol, traffic monitoring, biomedical, weather forecast, weapon imaging guidance,military reconnaissance and so on. Thus, it has wide application prospects and researchsignificance in studying objects tracking in dynamic image sequences. However, due toobject move speediness, complex background, illumination change, color approximatelybetween object and background, camera motion, et al, object tracking becomes verydifficult. Aimed at these problems, this research is mainly focused on the detection andtracking moving object in dynamic image sequences, the main work of which is asfollows:First, the key technology and prerequisites of moving object tracking is objectsegmentation. Two kinds of object segmentation algorithms were presented: One, formoving object detection in complex background, video object segmentation algorithmbased on minimum Tsallis-cross entropy is presented, which combined with framedifference detection to distinguish between the foreground and the background uses themethod of improved minimum Tsallis-cross entropy to set up the adaptive threshold.Experimental results show that the algorithm can overcome traditional shortcomings ofminimum Tsallis-cross entropy method, and can detect moving object more accurately incomplex background. The other, for moving object with speediness detection, videoobject segmentation algorithm is proposed based on marker multi-measure watershed,which achieves marker for gradient image through frame difference detection, andwatershed segmentation is used to acquire fast video object with precise boundary.Experimental results show that the algorithm can overcome the defect that it is easy toengender the excessive segmentations of the conventional watershed algorithm andobtain accurate boundary of the fast moving object.Second, two kinds of object tracking algorithms were presented in dynamic image object tracking. One, for tracking moving targets are susceptible to noise and trackingtime is too long, improved hausdorff video object tracking algorithm is presented, Withmoving object segmentation result as the initial template, it reduces matching timethrough motion estimation and removes random noise through transforming the binaryimage of the template and target image into multi-value image. Experimental resultsshow that the algorithm can overcome the defect that it is sensitive to the random noiseand long response time, and can track moving object with effectively and speediness. Theother, for traditional particle filter based on color error tracking in complex background,illumination change, color approximately between object and background, et al. Theparticle filter tracking algorithm based on multi-information fusion was proposed, whichcombines structural information and scale invariant feature transform (SFIT).Experimental results show that the algorithm can overcome the defect of traditionalparticle filter based on color in some degree, improving tracking stability and precision.Finally, study the experiment of fish swimming tracking based on object trackingalgorithm and set behavior parameters of fish swimming, achieving the purpose of waterpollution detection though analysis curve of motion characteristics parameters.Experimental results show that the statistical curve of motion characteristics parametersfrom the dynamic image processing can evaluate water quality and give early warning,thus protecting water environment.