Fast Compressive Object Tracking Algorithm Research Using Feature Weighting

Object tracking has important research value,and is widely used in video surveillance,intelligent transportation and many other scenes,and in recent years it has been a hotspot in the field of machine vision.Aiming at the problem of object tracking,many algorithms have been proposed.Although these tracking algorithms are robust,the visual target tracking is still a difficult problem due to the influence of external factors such as illumination and occlusion in the tracking process.As a research hotspot in the field of signal,compressive sensing has been widely used in tracking field because of its simple,efficient and real-time advantages.It is proved from the research that the low-dimensional compression signal obtained by the compressive sensing can basically preserve the whole characteristics of the original signal,which can greatly improve the computing efficiency under the premise of ensuring the accuracy of the calculation.The object tracking algorithm based on compressive sensing is simple and robust.However,when the multi-scale representation of the image is obtained,the effective feature generated by the filtering is gradually reduced with the increase of the filter size.At the same time,when calculating the similarity between the target and the candidate samples,the algorithm only simply superimposes the result of each weak classifier,and the processing strategy is too simple.When the appearance of target changes due to the influence of illumination,occlusion and other external factors,it is easy to reduce the tracking accuracy and make the tracking results appear deviation.Aiming at the above problems,this thesis proposes a fast compressive object tracking algorithm based on feature weighting.The algorithm achieves tracking by generating accurate high-dimensional features and using a more efficient similarity measure mechanism.The specific work of this thesis is as follows:(1)This thesis studies the theory of compressive sensing,and deeply understands the sparse representation of the signal,the design of the observation matrix and the common classification,signal reconstruction algorithms,compressed sensing dimensionality reduction,etc.Through the experiment,the validity of the compressive tracking algorithm is been proved,and the shortcomings of this kind of compressive tracking algorithm are also been found.Aiming at the problems,this thesis analyzes and resolves them.(2)According to the size of the filter,the adaptive weighting strategy is proposed to extract the multi-scale feature of the sample,and then to obtain the high-dimensional feature description.This method can solve the problem that the effective of the obtained feature decreases with the increase of the filter size,and ensures the accuracy of the use of the feature during tracking.(3)Considering the superposition of the result of each weak classifier and possibility of candidate sample compression features as the target,an efficient similarity measure mechanism is proposed.This method solves the problem of the single way to measure the similarity between sample and the target.In this thesis,we compare the proposed algorithm with the existing compressive tracking algorithms in different test data sets.The experimental results show that the proposed algorithm has higher accuracy.