Research And Design Of Moving Target Recognition And Tracking Method

With the rapid development of the transportation industry,the transportation system and its safety become more and more serious.The traditional artificial way is clearly unable to meet the current market demand.In this paper,A method of identification and tracking of ships and vehicles is studied.A target detection method based on mixed Gauss background and three frame difference is proposed,to extract potential area.The complete connected domain is detected by Two-Pass method and the line,shape factor and Zemike in the smallest external rectangular region are extracted.A LS SVM classification is designed to recognize and classify the object.After obtaining the initial position of the interested target,this paper mainly studies the identification and tracking of ships moving targets,we propose a target detection method of mixed Gauss background model and three frame difference based on the movement of potential areas,to complete the connected domain detection and extraction.The target in each frame of the video in the location,size and velocity information,are obtain by improved Camshift tracking algorithm for ship objects.There are many problems such as the abrupt change of illumination,the jitter of the camera and the fluctuation of the ripples in the video image.In this paper,we propose a hybrid Gauss background model and a three frame difference phase separation method,which can be used to obtain the potential active region by using the advantage of the classical mixed Gauss background model.And the connected domain is analyzed and merged to obtain the smallest outer rectangle,the geometric features of the target are extracted from the rectangle,and the LS SVMclassifier is constructed.In this paper,the average accuracy of target recognition method is improved obviously,the accuracy rate of ship is 95.45%,and the accuracy rate of vehicle is 96.15%.To resolve the problems existed in the estimation of motion in water video sequences,such as light mutation,shaking camera and undulating ripples,a target detection method based on mixed Gauss background and three frame differences is proposed to extract potential area.The minimum enclosing rectangle is obtained by analyzing and merging connected domains.The geometric features of the target are extracted from the rectangle,and the LS_SVMclassifier is constructed.Compared with two classical methods,this improved potential region detection algorithm can achieve the target detection in complex scenes accurately,which reduces the difficulty of the next recognition.Aiming at the problems that the current tracking algorithm may lose the ship objects,under the complex environment such as partial occlusion and the same color background,an ship tracking algorithm based on multi-feature Camshift and Unscented Kalman filter is proposed.The motion feature is extracted by the three frame difference,and fuse with hue and edge feature.Then the location of ship is calculated through Camshift algorithm.And Unscented Kalman filter is designed to predict and correct the center and size of the tracking frame.Bhattacharyya distance is also introduced to describe the degree of occlusion.Experiment results show the improved algorithm can achieve accurate and robust tracking under the complex situation,such as partial occlusion and the same color system background.