Research On Target Tracking Algorithms Based On WSN

With the rapid development of wireless sensor networks,computer vision and machine learning,target tracking technology has been widely used in in today's industrial production,environmental monitoring and daily life.In recent years,because the correlation filtering tracking algorithm has performed well,it has become a hot research object in the current tracking field.However,due to the complexity and variability of the scene environment in the process of target tracking,there are still a lot of difficulties to be solved in tracking technology.In this thesis,based on the correlation filter tracking algorithm is deeply studied,a series of improved schemes are proposed.The main research content is as follows:Firstly,aiming at the problem that the tracking effect of kernel correlation algorithm is not good when dealing with the target deformation,rapid movement of target,an improved target tracking algorithm based on heterogeneous feature fusion is proposed in wireless sensor network environment.Combining the excellent processing ability of template features for illumination change with complex background,statistical features are introduced as a new describing factor to make the global color features obvious and enhance the processing effect of fast movement and deformation,The tracking performance of kernel correlation tracker is further improved.The comprehensive performance of the algorithm is verified by simulation experiments.Secondly,aiming at the boundary effect problem caused by cyclic matrix in correlation filter tracking and the fact that the improved kernel correlation algorithm cannot realize the defects of scale self-adaptiveness,the spatial regularization method and the four-block method are studied.The addition of spatial regularization makes the training contain more negative samples,which reduces the inaccuracy of the sample,and solve the boundary effect caused by cyclic matrix.The four-block method calculates the scale of the current frame target indirectly by using the scaling coefficient.The comprehensive performance of the algorithm is verified by simulation experiments.Finally,aiming at the practicability of the improved algorithm,a real-time road monitoring and tracking system is designed,which combines three-frame difference and Mixture Gauss Modeling,and integrates them with the improved tracker in the wireless sensor network environment.The purpose is to detect the number of vehicles in a certain section of a certain period of time,and to complete the detection and tracking of vehicle information.The comprehensive performance of the system is verified by experiment.