Research On Object Tracking Algorithm In Video Compressed Domain Based On Particle Filter

In the field of computer vision,video moving object tracking has always been a research hotspot and difficulty.With video sequences to track people's moving objects of interest,this technology is widely used in video conferencing,video retrieval,intelligent monitoring,and pattern recognition..Visual target tracking can be defined as acquiring the motion parameters of the target to estimate the trajectory of the moving object.The operation domain of the tracking algorithm is divided into two categories: pixel domain and compressed domain.Algorithms in the pixel domain need to process massive amounts of information,and the data will limit its real-time performance when processing several videos in parallel.The algorithm in the compressed domain uses compressed data in the video bitstream,and its data information is block-based.Therefore,the amount of information in it is much less than the amount of pixel-based information in the pixel domain,and the calculation overhead when performing video information processing can be greatly reduced.In actual video applications,videos are compressed in different formats,and are often processed with useful information such as partially decoded motion vectors,block encoding modes,and residual transform coefficients.Compared to the pixel domain,the video is completely decoded and reprocessed.From the perspective of information processing,the computational overhead of processing video information is greatly reduced.This paper mainly studies the moving object tracking algorithm in the video compression domain based on particle filtering.The main work is as follows:1.Analyze the code stream structure of H.264 video coding,extract the vector features and macroblock layer information that can represent the target motion information,and pre-process the motion vectors to obtain the motion vector field that is most similar to the real motion of the moving target.2.Design a compressed domain target tracking method based on particle filtering.The algorithm uses motion vectors and block coding modes,and uses temporal and spatial correlations as system observations to estimate the state of the target,which can achieve moving target tracking in scenes with near uniform motion and background changes.Theexperimental results show that the comparison with the space-time Markov model in the compression domain shows that the tracking accuracy is greatly improved,and the comparison with the particle filter algorithm in the pixel domain shows that the processing speed is greatly improved.3.A particle filter tracking method based on salient region detection is designed to deal with the problem of target sudden motion tracking.The significance area is obtained by using the motion vector entropy and the smoothed residual norm.When the target moves suddenly and the tracking is lost,it can be updated and positioned in time in the saliency area to the target position.This algorithm improves the problem of target loss caused by sudden motion.