Research On Target Tracking Algorithm Based On Distributed Rao-blackwellized Particle Filter

Video target tracking has been widely used in video surveillance,intelligent navigation,intelligent transportation and other fields.Designing a robust video target tracking algorithm is a challenging task.In camera networks,the distributed particle filter algorithm is particularly suitable for large-scale nonlinear and non-gaussian distributed systems,but there are some limitations in the application of the camera network with limited communication resources and node resources,so it is necessary to further study the specific implementation of distributed particle filter algorithm.In view of the nonlinear,non-gaussian application scenario,this thesis proposes a RaoBlackwellized particle filter target tracking algorithm based on consistency algorithm,this algorithm first will be the product of the partial probability density approximation of each camera as the gaussian distribution,average consistency algorithm can be used to calculate the parameters of the gaussian distribution,so as to realize the tracking of the target state estimation.The specific work is as follows:(1)In a distributed network of cameras for target tracking the computing load and the data transmission of large amount of problems and the problem of tracking accuracy is not high,this thesis adopted in a distributed network of cameras in the camera run local particle filter method based on color features and add fusion filtering method based on consistency of distributed.Concrete implementation way of the method is all camera nodes in the network run respectively two kinds of particle filter,one is a kind of local particle filter based on color feature,camera nodes according to their own local filtering algorithm of target of observation,the second is fusion filter,used to calculate the global filter.This method distributes a large amount of computation to each camera node without fusion center and without a large amount of communication.The addition of color features improves the robustness of tracking.The simulation results also prove that this method can reduce data traffic and achieve more stable tracking,and the effectiveness of this method is verified.(2)For the camera due to angle,light and other problems caused by the target tracking failure,this thesis integrates multi-camera multi-angle tracking data,and combines the RaoBlackwellized particle filtering method based on color features with the distributed algorithm based on consistency.The distributed algorithm based on consistency is used to reduce the data transmission amount generated in the tracking and improve the transmission efficiency.The Rao-Blackwellized particle filter method based on color features is used to improve the tracking accuracy and enhance the robustness of the algorithm,a distributed Rao-Blackwellized particle filter target tracking algorithm based on consistency is proposed,and the experimental simulation is carried out to verify the efficiency and availability of the proposed algorithm.