Target Tracking Method Of Optical Measurement Image Based On Correlation Filtering

In recent years,with the rapid development of TT&C in the area of aerospace,the optical tracking system has become an important part of TT&C due to its high precision,strong visualization and stable performance.The parameters such as the trajectory,attitude,and optical radiation characteristics of the aircraft can be obtained by processing the optical images which are recorded by the optical tracking system.These parameters provide important data support for the performance evaluation and fault diagnosis on the aircraft.Optical image target tracking is a key part of optical image data processing.Its main task is to estimate the trajectory of target in video frequency.Investigating efficient and accurate tracking algorithm for optical images plays a key role in improving the real-time ability and the precision of optical image data processing.In this paper,target tracking algorithms based on correlation filter,which has been received wide attention in recent years due to their superior tracking performance,is the main research object.In order to improve the correlation filter in terms of the accuracy and the robustness in complex scenarios,the background information learning and feature fusion approaches are investigated in this paper.The main contributions of this paper are summarized as follows:(1)The traditional kernelized correlation filter(KCF)utilizes very limited background information of training samples and can easily drift in case of complex environmental disturbance.In order to address this problem,a scale-adaptive multi-feature KCF tracking algorithm based on context-aware is proposed in this paper.This algorithm incorporates context into the filter training stage and set the filter response close to zero for the context patches.The number of negative training samples is increased and the discriminating ability of the classifier is improved.The strategies of multi-feature fusion and scale estimation are also derived.Experimental results demonstrate that the proposed algorithm can take full advantage of the background information around the tracking target and improves the tracking performance in terms of accuracy and robustness.The proposed algorithm can effectively handle complex challenging scenarios such as illumination variation,scale variation,motion blur,fast motion,out of view,and background clutter.(2)Considering the robustness property of deep features and the accurate localization property of shallow features,a correlation filter algorithm based on adaptive fusing deep features and shallow features is proposed for target tracking in this paper.The data augmentation techniques are used to increase the number of deep feature training samples,and the convolutional neural network is used to extract deep features.Two continuous convolution filters are constructed to train the deep feature and the shallow feature samples independently.The response results of two filters are adaptively fused by using a novel fusing strategy based on prediction quality measure.The experimental results show that the proposed algorithm can take full advantage of the complementary properties of the depth features and the shallow features in terms of robustness and accuracy.The proposed algorithm exhibits high robustness and accuracy in case of complex scenarios such as scale change,motion blur,fast motion,and in-plane rotation.Both OTB dataset and optical tracking dataset are used to test the performance of tracking algorithms in this paper.The experimental results show that the proposed two algorithms both achieve higher accuracy and success rate than other related algorithms,indicating that the proposed algorithms can play a certain role in promoting the technical improvement of the aerospace optical tracking measurement.