Building A Key Frame Template Dictionary And Online Learning For Robust Offline Video Tracking

Object tracking,which rises in the last century,is an emerging technology integrating many disciplines.It has promising applications in various academic fields,especially in the field of computer vision.Scholars have always regarded it as the basis of visual tracking technology and the following high level analysis.So the ability to accurately tracking the target is crucial,which is the main reasonfor most scholars to study the subject.Simply speaking,object tracking constructs the matching target chain from the image sequence using the time and space consistency of two adjacent frames.In a century,object tracking algorithms emerge in an endless stream.Studying a lot of literature,we find that a large number of algorithms can only meet the requirements of specific environments,which is different algorithms can only achieve satisfactoryresults for a particulardata set.This thesis presents two algorithm.One applies to tracking frame.It is an improvement of the tracking based on sparse representation.The proposed algorithm improves tracking mode,construction of template dictionary,and computational efficiency.Another is an improvement of appearance model.The algorithm combines the spatial and temporal distribution and extends the single timing tracking to spatial and temporal joint distribution tracking.What two algorithms have in common is that they are the further improvement of the original and make the performance and scope of tracking more wide.Tracking simulation is made based on the proposed algorithm usingvarious data sets.Experimental resultsdemonstrate the wellperformanceof the proposedtracking algorithm.