Research On Moving Target Detection Algorithm For Surveillance Video

In recent decades, with the advancement of science and technology, the vision field of imageprocessing with the core computer has been considerable developed. As one of the core researchtopics in the field of visual image processing, moving target detection, has been widely used invarious fields, such as intelligent surveillance, medical and military. Therefore, moving targetdetection algorithm is of important theoretical significance and application value. This papermainly analyzes the moving objects video frames which are taken by static videos. Comparedwith some traditional methods, this paper makes some improvements, and does simulationexperiments to confirm the effectiveness of the improved algorithm.Firstly, this paper makes a brief introduction to moving target detection related theories: theimage pre-processing, mathematical morphology processing, edge detection, and so on. Thenexplores the existing three classic moving target detection algorithms: the frame differencemethod, optical flow method and background subtraction method. The principle, the advantagesand disadvantages of the three algorithms are analyzed in detail. And also discusses theprinciples of the background subtraction method using Gaussian mixture modeling and thethree-difference methods.Since the traditional Gaussian mixture model can not adapt well to the change of light, thispaper provides different parameter update program from the local illumination and globalillumination mutation respectively; Due to the noise, the object exists broken "empty" and alsothe boundary contour is not clear enough by the traditional three-difference method. Therefore,this paper proposes a new idea: three-difference algorithm combined with Kirsch edge detectionoperator. Finally, Obtain the result by logic algorithm of the improved Gaussian mixture modeland three-difference method as data fusion, and does simulation experiments using OpenCV inVisual C++6.0platform. The results show that the new fusion algorithm could extractclearer and more complete moving object image, and show better adaptability and robustness forindoor or outdoor video sequences.