| Video object tracking has been widely used in various fields of military andcivilian affairs, such as video surveillance, robot navigation, human-computerinterface, safety inspection and traffic management etc. This thesis focuses on theapplication of the Mean Shift algorithm in the object tracking under theenvironment that the camera is stationary with moving object. Some improvedobject tracking methods are presented in this thesis, which will contribute topromoting the tracking accuracy of moving object and have better stability.Owe to its advantages of real-time, robustness, Mean Shift method, as one ofthe most popular methods for object tracking in recent years, has drawn a highdegree of attention and been widely used. However, traditional Mean Shift objecttracking algorithm has its own limitations, such as its impossibility of consideringthe spatial information and its inability to meet the target diversity based on thesingle characteristic description. As for these defects, this dissertation presentssome corresponding improvements. Several aspects around these key problems willbe studied in this dissertation. The main work of this dissertation is as follows:1、Firstly the basic theory of Mean Shift algorithm is overviewed, such as itsresearch status and its convergence. By analyzing the physical meaning of the backproject image and the superiority of the histogram in describing moving object,then this dissertation presents an optimized method of Mean Shift algorithm.2、Traditional Mean Shift algorithm calculates the back project image only bythe similarity of single point. As for this defect, an improved method based ontemplate similarity is given, which promotes the tracking robustness.3、Traditional Mean Shift algorithm only uses the color characteristics todescribe an object, and the detection precision is not good enough. As for thisdefect,we combined the texture characteristics,together with color feature(HSV)topromote target tracking accuracy. |
PDF Full Text Request