Research On The Application Of MeanShift Algorithm In Fish Tracking System

Bionic fish is a typical class of underwater vehicles,compared with traditional underwater vehicles,there is greater flexibility and broader application areas.In order to study the propulsion principle of bionics fish,the tracking trajectory of the bionic fish can be tracked by the object tracking method of machine vision to obtain the specific position coordinates.In this paper,the basic principle of target tracking is studied,the MeanShift algorithm,Calman filter algorithm,the particle filter algorithm are compared.finally,the MeanShift algorithm which is representative of the non parametric estimation is chosen as the basis of bionics fish trajectory tracking.The histogram of the target region is used as the tracking feature to track the target position,less computation,not sensitive to the deformation and rotation of the tracked object,and it has a high accuracy in tracking moving objects in low speed.In order to achieve the purpose of accurately tracking the trajectory of the bionic fish,this paper uses the MeanShift algorithm based on the combination of theory and experiment.First of all,considering that the traditional MeanShift algorithm in tracking fast moving objects is not accurate,loss of the search box and other issues,put forward to improve the MeanShift algorithm based on diamond search algorithm in motion estimation,the mechanism of adding prediction process was iterative in the tracking algorithm,first from the predicted candidate region in search,avoid the complex process of global search,reduce the computation time of the algorithm,to solve the problems of fast object tracking.Secondly,this paper simulates the operation and MeanShift algorithm is built in Simulink xPC-Target platform,the host and target machines interconnected way to solve the problem of real-time motion tracking,used for bionic fish swimming video capture camera connection on the target machine to connect the TCP/IP video information is uploaded to the host the host,used for video image processing program in the host computer,processing after the program is downloaded to the target machine can separate video information.Finally,in order to realize the trajectory tracking of 3D bionic fish swimming,the binocular vision,the double target has left and right camera internal and external parameters by least squares method,three-dimensional measurement,known at any point in space on the left and right camera coordinate coordinates,calculated by the solution of homogeneous equation in three-dimensional space coordinates of the point.In order to verify the feasibility of the above method,the innovation points are confirmed by the experimental method.First,the traditional MeanShift algorithm and the improved algorithm of motion estimation and tracking method with contrast tracking experiment,as tracking video with bionic fish swimming video speed change,compared with the tracking method and the improved traditional Mean Shift tracking algorithm,we can see that the improved method can effectively solve the problem of fast moving time in the target tracking accuracy and tracking does not appear frame loss problem,illustrate the feasibility of the algorithm.Second,the establishment of Simulink model in the establishment of good xPC-Target platform,image processing program to achieve the goal of video acquisition and host download,bionic fish swimming video camera in real time using two case collection,will collect information uploaded to the host after the improved MeanShift algorithm is used for the programming.Real time display in the host location coordinates of bionic fish swimming.The real swimming trajectory of the fish is compared with the tracked trajectory,and the accuracy of the tracking is illustrated.Third,in view of monocular camera can only be realized in the xPC-Target platform model for target tracking,trajectory tracking two-dimensional,in order to more accurately simulate the swimming trajectory of bionic fish,we introduce binocular camera video capture,camera through the coordinate system,through the use of the least squares method,get the coordinates of each point track,and draw the trajectory of three-dimensional swimming robot fish swimming.The real 3D swimming trajectory of the fish is compared with the tracked 3D trajectory,which shows the accuracy of the tracking.It can be concluded that the bionic fish trajectory based on improved MeanShift algorithm based on binocular tracking system combined with the camera in the xPC-Target platform,can be two-dimensional or accurately bionic fish swimming trajectory in 3D space target tracking,tracking experiments show that the bionic fish swimming trajectory trajectory and actual in a small error and it shows that the system has high accuracy.