An Anti-occlusion Of Object Tracking Based On Kernelized Correlation Filters

With the rapid development of Artificial Intelligence,the identification of video objects and the interpretation of in-depth information had become a research in smart monitoring,visual navigation,intelligent transportation,human-computer interaction and national defense reconnaissance.The target tracking technology is the basis of video target recognition and interpretation.The core of the target tracking algorithm is to have high accuracy and robustness.Extracting the HOG features of the initial frame observation object,the Kernel Correlation Filters(KCF)target tracking algorithm constructs a large number of feature samples using cyclic shift,and uses the kernel ridge regression method to train the classifier to predict the target position with accuracy.It is highly popular for its high speed,high speed and high robustness.However,in practical applications,when the target with the background clutter,scale variation,deformation,illumination variation,motion blur and rotation,et al.the KCF algorithm relying on a single HOG feature is easy to cause the target to lose.At the same time,the observed target is often affected by other static or moving objects.The lalgorithm with handling is a challenging subject that few people are involved in.The paper takes the practical application requirements of the research project as the traction,and devotes itself to the use of nuclear related filtering target tracking framework theory to explore the construction target feature expression model and multi-feature fusion rules.The anti-occlusion KCF target tracking algorithm under complex scenes with background clutter,scale variation,illumination variation,deformation and the target being occluded is studied to improve the accuracy and robustness of KCF target tracking in complex scenes.The main research contents of the paper are:(1)The defects of the traditional KCF target tracking algorithm are studied.The multi-feature fusion rules are proposed and the multi-feature fusion KCF target tracking algorithm is implemented.The validity and accuracy of KCF target tracking depend on how to reasonably establish the target feature model according to the characteristics of the target scenario.The paper firstly analyzes that the traditional KCF algorithm relies on a single HOG feature,scale variation and deformation significantly,it is easy to lead to the loss of target tracking.On this basis,based on the characteristics of the scene,a multi-feature weighted fusion rule that takes into account the characteristics of the target HOG,LBP,and CN is proposed,and a multi-feature fusion KCF target tracking algorithm adapted to target tracking tasks in complex scenes is implemented.(2)KCF anti-occlusion target tracking algorithm with partial occlusion is proposed.Based on the research of the multi-feature fusion KCF target tracking algorithm,for the non-complete occlusion target tracking task,the current frame and the previous frame peak value are calculated according to the method of solving the peak sidelobe ratio of the output response map of the adjacent frame nuclear correlation filter.The average value of the sidelobe ratio is used as a threshold for judging whether the observed object is occluded in the current frame: if the ratio of the peak sidelobe of the current frame is smaller than the current frame threshold,the current frame is deemed to have been occluded by the observed object,and the current frame kernel is not updated.Corresponding filter model,an adaptive target occlusion judgment method and an anti-occlusion kernel related filter target tracking algorithm are proposed.The comparison experiments show that the proposed method has better tracking accuracy and robustness when the target is partially blocked and severely blocked.(3)An effective tracking method in the case of a complete target occlusion is proposedBased on the proposed partial occlusion adaptive decision method and anti-occlusion KCF target tracking algorithm,aiming at the short-time complete occlusion problem of the target,when the observation target reappears,the combination of naive Bayesian and KNN combined classifier can effectively achieve the reappearance of the observation target to realize the KCF tracking algorithm under the condition that the target is completely occluded for a short time.The comparison experiments show that the proposed method has better tracking accuracy and robustness when the target is partially blocked and severely blocked.