Research On Visual Object Tracking Algorithm Based On Space Penalty And Correlation Filtering

Visual object tracking can solve the problem of object trajectory positioning.It has been widely used in intelligent security,intelligent transportation,unmanned driving and biomedicine fields.The emergence of a large number brilliant talents and the injection of multiple funds have promoted the development of visual object tracking technology.And many excellent visual object tracking algorithms have emerged in a short time.However,problems such as object tracking under complex backgrounds,long-term object tracking,real-time object tracking,adaptive multi-feature selection,and adaptive object scale estimation cannot be properly solved.This paper is based on the spatial target and correlation filtering visual target tracking algorithm,combined with the SRDCF(Learning Spatially Regularized Correlation Filters for Visual Tracking)algorithm in this field,and conducts in-depth research on robust target tracking methods in complex environments.Based on the SRDCF algorithm,the speed of tracking targets is improved,and the accuracy of tracking targets under occlusion is improved.The main work and results of this article could be concluded as follows:(1)Study the possible improvement direction of SRDCF algorithm.The SRDCF algorithm is an improved algorithm based on the MOSSE and DCF algorithms.First of all,it is only violently divided into several specified scales and then matched in sequence.On this basis,we can carry out research on adaptive scale or super-large scale.This tracking effect will be further improved.Secondly,when solving the correlation filter,it uses the Gaussian iterative method for online learning.The Gaussian Sider iterative algorithm can be replaced with the ADMM(Alternating Direction Method of Multipliers)iterative algorithm to improve the tracking speed.Finally,occlusion detection can be introduced to further improve tracking accuracy.(2)The SRDCF algorithm is accelerated and improved based on the ADMM iterative algorithm.Firstly,the object search area is expanded,which is 25 times that of the target area.Then the ADMM algorithm with faster solution speed is used to replace the original Gaussian Sidel iterative method,and the tracking speed is accelerated on the basis of no loss of tracking performance.On 100 complex video sequences with multiple attributes,perform overall performance evaluation and performance evaluation of each attribute.The overall performance evaluation experiment results show that the ADRD-based accelerated SRDCF algorithm studied in this paper not only achieves acceleration,but also improves the tracking accuracy.Through the experimental results of the performance evaluation of each attribute,it is found that the algorithm proposed in this paper can cope with a variety of complex tracking environments under the premise of real-time tracking of 26 FPS.(3)Improve the SRDCF algorithm based on occlusion detection.The improvement idea is to add tracking quality detection and subsequent processing on the basis of SRDCF.The joint index of PSR(Peak Side Lobe Ratio)and maxF is used to determine whether the target is blocked,and the filter is not updated when the block is detected,which can prevent the filter from being polluted by invalid targets and causing tracking failure.From the experimental results,the SRDCF algorithm based on occlusion detection proposed in this paper(hereinafter also referred to as the anti-occlusion SRDCF algorithm for short)is superior to the SRDCF algorithm in overall performance evaluation,and also superior to the other three comparison algorithms.