Optimize Detection Research Of Video Moving Object

Video moving object detection technology is an basis technology of intelligent video surveillance system,and its main tasks are: to detect moving object of attention from the video sequence,to obtain moving object information as accurate and complete as possible,the test results will affect the subsequent video surveillance system technology.In order to improve the performance of video moving object detection,this paper mainly launches the research from the weighted Kolmogorov-Smirnov algorithm and dynamic graph cut algorithm in two aspects,the weighted Kolmogorov-Smirnov algorithm is designed to improve the similarity and accuracy,dynamic graph cut algorithm is designed to improve the efficiency.Gaussian mixture model is often used in video moving object detection with a complex background of a slight disturbance,such as shaking leaves,water fluctuations.Due to the influence of the external environment or human factors,normally this background disturbance is unstable,even though the Gaussian mixture background model is continuously updated,in order to improve the accuracy of detection,optimized the weighting coefficients of background model is still very necessary.Also,because the estimation of the weighted Kolmogorov-Smirnov algorithm has consistency and robustness,therefore,this paper proposed video moving object detection based on the weighted Kolmogorov-Smirnov algorithm.The main contributions are: 1)the algorithm theoretically and experimentally proved that under the conditions of Gaussian mixture model,the estimation of the weighted Kolmogorov-Smirnov algorithm has consistency and robustness,indicate the weighted Kolmogorov-Smirnov algorithm can more accurately adjust the weighting coefficient of Gaussian mixture model.2)For non-stationary background disturbance problem,single model of the object detection ability is limited,this paper cascaded the weighted Kolmogorov-Smirnov algorithm and Gaussian mixture model together,on the one hand the weighted Kolmogorov-Smirnov algorithm can accurately adjust the weighted coefficient of Gaussian mixture model;on the other hand,Gaussian mixture model can provide initial moving object area for weighted Kolmogorov-Smirnov algorithm in object detection.The simulation results show that the proposed algorithm has high detection accuracy.Graph cut algorithm based on energy minimization method is widely used in image processing and computer vision,but when it segment image,it is need to establish a network graph for each image,and the difference of adjacent two images is not big,thus increases the number of repetitive tasks,reduces the computational efficiency of the algorithm,which limits its further development.To address this problem,the paper carried out research on the dynamic graph cut algorithm.The algorithm established Gaussian mixture background model,it can be applied to video sequences with complex dynamic background in object detection.The algorithm is also based on the features that the pixel data gap of two adjacent frames of a video image is small,only to establish a network graph of the first frame image.In order to avoid duplicate calculations,dynamic graph cut algorithm gives a maximum posteriori probability calculation of specific dynamic Markov Random Filed,and the remainder obtained by updating the edge flow of previous frame network graph,making it characterized the pixel date of the current image frame.In this update process,in order to maintain the consistency of the edge capacity,dynamic graph cut algorithm gives the update method that has consistency.Compared to graph cut algorithm that each frame images are to establish a network graph,the efficiency of dynamic graph cut algorithm has been greatly improved.