Research On Moving Object Tracking Algorithm Based On Particle Filter And Local Sparse Representation

Moving object tracking is an research hot spot in field of computer science.It's purpose is to estimate and predict the status of the moving target through using the modern computer information processing technology,and make the machine has the ability to track the moving object in the video image sequence.Then,provide a reliable basis for further target behavior analysis and understanding.At present,great progress has been made by both academic and industrial's in-depth research on the moving target tracking theory and technology development.However,due to the complexity of the moving target tracking problem itself,design an accuracy,robustness and with consideration of real-time moving target tracking algorithm still a challenging problem.In this paper,with the application background of moving target tracking,a study of the particle filter theory was performed.In order to meet the needs of the tracking accuracy and robustness of the moving target in complex environment,a research on the improved particle filter itself and the specific application on moving target tracking were proposed.And carry out the theoretical analysis and experimental demonstration.The work of this paper are as follows:Particle filter presence of particle improverishment and particle degradation.To solve these problems,an improved particle filter scheme was put forward based on artificial fish swarm algorithm,which the optimization mechanism has the characteristics of randomly and the search direction of the diversity.By introducing it's preying and gathering behavior into the particle filter resampling procedure,it can induced particle movement in high likelihood region.Based on this,the thesis also analyses the artificial fish swarm algorithm is easy to fall into local extremum with the premature convergence phenomenon,and introduced another chaotic artificial fish,further improved particle filter algorithm.The experimental results has show that,the improved particle filter algorithm can effectively avoid the particle degradation and particle improverishment problem.In the calculation of overhead is not significantly increased,it can significantly improve the filtering accuracy.In the practice of moving object tracking,the particle filter is easily affected by factors such as illumination change,appearance change,occlusion and background clutter.To establish a static impression and a dynamic model of the object's appearance,a sparse coding method based on local image block is designed.In addition,the aim of the object tracking is posed as a binary classification problem for discriminate the object target from the background image.A two-stage particle filtering method was used to alleviate the tracking drift problem caused by model update.The adaptive algorithm can capture the target appearance changes and reduce the errors caused by the wrong location update.Experiments has demonstrate that our proposed algorithm can adapt to the environment changes.And the tracker can overcome the global template matching flexibility in processing the target local variation.The accuracy and robustness were improved.