| Object tracking not only plays an important role in robot vision,intelligent monitoring,unmanned driving and so on,but also is one of the indispensable technologies to accurately guidance of missile weapons.Therefore,it is of great significance to study the object tracking algorithms with high accuracy,strong robustness and good real-time performance.This paper summarized the progress of object tracking in recent years at home and abroad,and on this basis,it introduced two directions of tracking algorithms at present:correlation filter and deep learning,and carried out application research on them.The research results showed that on the one hand,the use of hog feature in correlation filters can effectively improve the tracking accuracy,but the generation process of the feature descriptor is complex,which greatly reduces the tracking speed;on the other hand,the use of deep learning can deal with more complex tracking scenarios,but the pre-training of neural network model depends on a large number of data and hardware GPU support,which can not meet the requirements of guidance missile‘s flight operations.In order to solve the above problems,this paper proposed a feature extraction method based on single-layer perceptron model,named the obtained features as AOB feature,and transplanted the AOB feature to correlation filter algorithm dsst for test,and gave an objective analysis and evaluation of the experimental results.The evaluation data showed that the tracking speed of the algorithm using AOB feature is more than twice as fast as that using hog feature,and the tracking accuracy is also ideal.In addition,this paper also took two operational targets of missile weapons as the testing objects,they are moving tank and the solid interceptor entering the water.The tracking performance of AOB feature was basically consistent with the hog features,which verified the feasibility and effectiveness of AOB feature in military applications.Simulation experiments showed that the AOB feature proposed in this paper have the advantages of simple extraction method,no use of color features,no training,and strong portability.It is suitable for TV seeker guidance technology and also provides a new idea for feature extraction of object tracking. |
