Research On Target Tracking Algorithm Based On Particle Filtering With PCA In Complex Environment

Particle filtering is one of the mainstream technologies u sed in target tracking,however,it is always a challenge to achieve effective tracking of targets in complex environment.In order to solve the problem of feature space failure caused by environment change,this paper has carried on thorough research on the particle filter technology with principal component analysis(PCA).The main research work and innovation are as follows:(1)A new particle tracking target tracking method based on PCA is proposed.The problem of updating the feature space in complex environment is solved by using quadratic PCA.The front-end PCA decomposes collection of target templates to establish the target feature space,in the following tracking,the particle vector is projected into the feature space to calculate reconstruction error,and then the state of the target in the next frame is estimated by the weighted sum of the particles.Our algorithm can achieve fast and effective effect of feature space updation.(2)A judgment method of feature space updating based on Lorentz estimator is proposed to avoid frequent updates reducing the processing effect of the problem.The estimator applied before the second order PCA makes full use of the exclusionary effect in the process of parameters estimation to obtain the number and distribution of outliers which caused by changes in the appearance of the target or occlusion,so we can statistic the number of outliers to determine the current frame updating time.(3)A multi-target tracking algorithm based on agent mode is proposed.The algorithm assigns a tracker to each target independently,and adds an information sharing module to handle the occlusion between the targets.When the occlusion occurs,it first judges the occlusion type by analyzing the information in the shared module,and then using Kalman Filter to predict the moving direction of the target,expand the search range,and finally estimate the target possible location with weighted search results.The algorithm can effectively deal with the occlusion problem between multiple targets.The experiment and simulation of the above algorithm are carried out through the open monitoring video test sequence.The experimental results show that the second-order PCA can correctly update the change information of the target appearance int o the feature space.The updating judgment method based on the Lorentz estimator can greatly reduce the unnecessary update.The multi-target tracking algorithm based on agent mode can effectively deal with the occlusion between targets,and realize the effective tracking of multi-objective in complex environment.