Research And Application Of Anti-occlusion Object Tracking Algorithm Based On Correlation Filtering

Moving object tracking technology is a key research direction in the field of visual image processing.It is widely used in video surveillance,military guidance,human-computer interaction and other fields in the intelligent era.Among them,the traditional correlation filtering object tracking algorithm is widely concerned by researchers and has achieved remarkable research results because of its simple and efficient advantages.However,due to the complexity and variability of the tracking scene,the current tracking algorithm has different degrees of defects in practical application scenarios.Therefore,it is of great practical value to study how to achieve efficient and robust object tracking in complex situations.In this paper,the main line of research is to realize object tracking in complex scenes.Based on the framework of correlation filtering tracking algorithm,corresponding improvement methods are proposed to overcome the shortcomings of the algorithm in occlusion,fast motion,complex background and other tracking situations.The main research work of this paper is as follows:(1)Aiming at the inaccurate position prediction caused by occlusion,fast motion and other disturbances in the object tracking process,a long-term object tracking algorithm combined with Kalman filter prediction is proposed.Kalman filter is introduced in front of the re detection module to predict the object position in case of object loss due to occlusion and other disturbances,so as to avoid the traversal detection of the whole frame image by the re detection module;At the same time,PCA dimension reduction method is introduced to reduce the dimension of the scale prediction filter,and QR decomposition interpolation is used to reduce the calculation cost of the scale filter when the scale prediction accuracy is equivalent.The experiment on OTB-2015 dataset shows that the improved method can effectively predict the object position in interference scenes such as occlusion,and improve the tracking accuracy and real-time performance of the algorithm.(2)To solve the problem of unstable tracking of correlation filter tracking model in complex scenes such as background interference and occlusion,this paper proposes a object tracking algorithm based on spatio-temporal context awareness fusion re detection mechanism.Firstly,the spatio-temporal context awareness item is added on the basis of the filter model to improve the filter’s ability to classify the object and background by learning the spatio-temporal context relevance information of the object;Secondly,it is proposed to judge the reliability of object tracking according to the average peak correlation energy,and then decide whether to update the model to prevent the model from being polluted;Finally,the SVM re detection module is added to the algorithm tracking framework.When the predicted response of the object position is less than the set threshold,it indicates that the object tracking is lost,and the re detection module is used to relocate the object.The experimental results on OTB-2015 and Temple color 128 datasets show that the CARDCF algorithm proposed in this paper has better tracking performance and robustness than the mainstream tracking algorithm in complex interference scenes such as occlusion.(3)In order to realize the application of the improved algorithm in the actual tracking scene,the improved long-term object tracking algorithm is transplanted on the embedded system based on Raspberry Pie.Through the raspberry pie hardware platform,Qt and Open CV development environments are built to achieve the task of object image acquisition and object tracking.The experimental results show that the improved algorithm has strong robustness and real-time performance when tracking objects in the embedded environment.