Research On Target Algorithm Based On Improved Sparse Coding Particle Filter

Target tracking is one of the most important technology in intelligent monitoring systems and is the key to intelligent operation.Its accuracy is directly related to the accuracy of intelligent monitoring.The target tracking algorithm incorporates various ideas such as image processing,automatic control,machine learning.So it is widely used in the video field.However,due to outdated equipment in substations,many surveillance videos need to be manually viewed in real time,so the work of workers is low efficiency.Therefore,there is an urgent need for an algorithm that can achieve automatic tracking of targets in video.This article chooses particle filter target tracking algorithm to complete the project in many tracking algorithms.The particle filter target tracking algorithm estimates the posterior probability of the target motion through the collected particle samples,and selects the final position of the most target of the maximum particle sample of the posterior probability.The algorithm is widely used in video tracking because it is not restricted by the state space model.However,when the environment is complex,a large number of particles are needed to accurately estimate the target position.Therefore,the algorithm increases with the increase of the number of particles and the running speed is slow.Particle filter target tracking algorithm can not automatically select the target,so this paper chooses moving target detection algorithm based on the SURF feature matching to solve the problem that particle filter target tracking algorithm can not automatically select the target.For the particle filter target tracking algorithm slow tracking and the target can not automatically select the problem,the main research work of this paper is as follows:(1)A particle filter target tracking algorithm based on improved sparse coding is proposed.In this paper,the traditional sparse-coding target model is improved.This paper use L2 paradigm constrains sparse coefficients,L1 paradigm constrains trivial template coefficients.And Ridge regression and fast iterative threshold shrinking algorithms are used to iteratively solve sparse and trivial template coefficients of the target model.This new algorithm Effectively reduce the complexity of the target model and improve the solution speed of the target model.(2)A moving target detection algorithm based on spectral residuals and improved k-means clustering algorithm is proposed.This algorithm introduces spectral residual algorithm and k-means clustering algorithm into the target detection algorithm based on SURF feature matching.As a result it solves the traditional algorithm's problems.such as pseudo target feature point matching is not accurate and target contour is not continuous.(3)This article will integrate improved detection and tracking algorithms to achieve automatic tracking of the target and detect the effectiveness of the algorithm's integration in the actual captured video.In the above four works,we want to improve tracking accuracy and success rate of tracking algorithm by changing the target model of tracking algorithm and the method of solving the target model.By introducing spectral residual algorithm and k-means clusteringalgorithm,the detection accuracy of the target detection algorithm is improved.As a result,the problem of particle filter target tracking algorithm cannot automatically select the target is solved.