Moving Object Detection And Tracking Algorithms And Their Applications In Video Sequence

Dynamic object detection and tracking are the important issues of computer vision and thehotspot of the certain research domain. In recent years, the target detection and object trackinghave attracted many researchers’ attention. Although many effective visual object detection andtracking methods have been proposed, there are still a lot of difficulties in designing a universal,accurate, robust, and real-time tracking algorithm due to the challenging complex scenarios suchas significant illumination changes in environment, occlusion, pose variations of the object andnon-linear deformations of shapes, and noise and dense clutters in complex background, etc.Therefore, the researching on moving target detection and object tracking under complicatedbackground has both important theory significance and application value.The main research contents and contributions of this dissertation can be summarized asfollows:(1) Focus on the problem of target detection between traditional inter-frame differencemethod and background subtraction method in the case of complex background, illuminationchanges, an improved target detection method based on the combination of three-framedifference method and adaptive Gaussian mixture model is proposed under the traditional targetdetection algorithm (Inter-frame difference method, Background subtraction method andGaussian background model method). Firstly, this method dynamically build a backgroundmodel area using adaptive Gaussian mixture model through background subtraction method todetect moving targets. Then, adopting three-frame difference method to get the moving targetregion in two adjacent gray-scale images. Finally, a real moving target can be extracted bymovement target areas which use adaptive Gaussian mixture model for background subtractionand three-frame difference method. Because of combining the advantages of three-framedifference method and adaptive Gaussian mixture model method, the method can quickly createand update the background to overcome illumination changes and background noise andinterference, and accurately separate the foreground object.(2) A new Camshift target tracking method based on adaptive multiple features fusion isproposed to solve object model based on single visual feature not having enough robustnesswhen environment changing much. The algorithm build target model based on color and texturefeatures distribution, calculate the degree of similarity between the current frame the candidate model and the target model by Bhattacharyya coefficient, and calculate the adaptive fusionweights according to the contribution of the respective feature on the target model. The methodachieved adaptive target model updating and improved the tracking accuracy and anti-jammingcapability. In order to solve the problem of target tracking under occlusion, the algorithm uses athree-point forecast estimation method to handle occlusion in the tracking process whicheffectively solved the target morphological changes and partial occlusion phenomenon.(3) Because Camshift object tracking algorithm occurs template drift when similar targetsand background interference in the template updating, particle filter target tracking algorithm canbe introduced in this paper. Because particle filter can effectively solve the estimation problem inthe non-linear and non-Gaussian system, it has been widely applied to the field of object tracking.But the standard particle filter algorithm can not solve the problems of particle degeneration andsample impoverishment in object tracking. To solve these problems, an improved particle filterobject tracking algorithm is proposed, which based on dynamic adaptive evolutionary strategy.The algorithm dynamically computes the feature’s fusion weight by the discriminability of eachvision feature, and then constructs the important density function based on selecting featuresfusion method adaptively. Moreover, the self-adaptive genetic evolutionary mechanism isintroduced into the particle resampling process. With the self-adaptive crossover and mutationoperators, the new particles can better approximate the true state of tracking object. Compared toconventional particle filter, the new tracking object is very challenging regarding illuminationvariation, structural deformation, interference of similar target and occlusion.In the dissertation, we studied several key technologies under complicated environment inthe target detection and object tracking algorithms, such as movement target detection andsegmentation, target area features acquiring, target describing, and real-time, robust objecttracking, and proposed effective improved method. And the effective of improved targetdetection and object tracking method has been verified in the practical application.