Research On Moving Target Tracking And Abnormal Detection In The Video

With the rapid growth of the population and the rapid development of the city,the public safety has been becoming an urgent problem for the world.The massive unstructured monitoring data not only requires the effective storage and management,but also needs real-time data analysis and retrieval,the monitoring method based human has been unable to meet the demand,therefore the realization of intelligent video surveillance has significance for the maintenance of the public safety.This paper mainly carrys on an in-depth research in the video moving target tracking and anomaly event detection,proposed the RB particle filter moving target tracking algorithm based on the DPM model and the video anomaly detection algorithm based on graph structure,which details are as follows:In the moving target probability tracking algorithm,the unified state space equation is often used to describe the linear Gaussian relationship between the unknown state variables and the known observations.However,due to the complex climate,dynamic background interference,target complex movement and other factors,the state space equation often presents a nonlinear non-Gaussian relationship,so we need to choose the Bayesian filtering algorithm flexibly to estimate the state variables in the process of target tracking.To this end,this paper proposes a novel target tracking algorithm,the Rao-Blackwellised(RB)particle filter tracking algorithm based on the Dirichlet process mixture(DPM)model.In the nonlinear state-space equation with linear substructure,the DPM model is used to describe the distribution of the observation noise and to automatically adjust the number of Gaussian mixture model components based on the observational data,which improves the tracking accuracy.The RB particle filter algorithm estimates linear and non-linear variables using Kalman filter and particle filter respectively,which improves the computational efficiency.Improved accuracy and efficiency are helpful to the moving target tracking.The experiment on the simulated situation and the UCSD data set shows that this algorithm has good tracking performance in the moving target tracking.In the abnormal event detection based on statistical learning modeling,feature extraction is essential to the whole statistical model,due to the interference of obstacles in the complex scene and an isolated point that caused by noise in the optical flow field,this paper proposes a video anomaly detection algorithm based on graph structure,which intends to capture the underlying intrinsic structure of characteristic data.It uses a graph signal processing way to mining the strong coupling between the feature data,thus extracts the best feature data in complex scenes,and completes learning and anomaly detection of the joint space-time model.At the learning stage,ituses the method of weighted tree to learn model parameters.First convert the model into two spanning trees as i HMM-LDA and LDA-iHMM structure and learn each parameter of the structure,then weight the parameter to optimize the joint spatial-temporal model.When in the stage of abnormal detection,it calculates and compares the log likelihood function of the train videos and test videos via the forward-backward algorithm.The experiment on the public data sets UCSD and UMN verifies the excellent detection performance in the graph structure in the abnormal event detection.