Object Tracking Research Based On Compressive Sensing

The application of object tracking is widely used in compute vision,which is simultaneously a basic reasearch.In recent years,many new algorithm of object tracking is put forward,and the accuracy and complexity of algorithms make significant progress,however there exists many pracical scenarios where tracking algorithm can't meet the demand in many industrial areas therefore there is still need to improve this kind of technology.We summarizes the work and innovations as follows:Firstly,we introduces commmonly used method of feature extracion and the tracking theories in the video object tracking,which include the theory of compressive sensing feature extraction and machine learning used in tracking.Secondly,compensation of point tracking is used in kernelized correlation filters tracking algorithm for the lack of adaptability when used in tracking object with scale changes,the result of point tracking is used to compensate for the shift and scale changes,the feature extracted from compressive sensing is used to model the tracking object to prevent the error of mutation associated with kernelized correlation filter tracking brought from point tracking.Lastly,to solve the promblem of low accuracy in compressive sensening tracking,which often causes the failure of tracking in some scenarios,multi-feature tracking algorithm is used.It increase the stability of tracking when color space information,point feature compensation, variable rate learning is used.And experiment is conducted to verifiy the effectiveness of this method in practical applications.