Research On Multi-feature Fusion Target Tracking Algorithm Based On Correlation Filtering

At present,with the complex application environment and stringent technical requirements,the target tracking technology needs to be improved.Among them,feature extraction is an important improvement direction in tracking technology.From simple feature to complex feature,from single feature to multi-feature,we constantly optimize the expression ability of feature to the appearance of target.Based on the theory of correlation filtering framework,this paper focuses on feature selection in tracking process.At the same time,considering the real-time and robustness requirements of the algorithm,combined with a variety of correlation filtering target tracking algorithms to carry out research work.The specific research contents and innovations are summarized as follows:(1)Aiming at the problem of tracking drift or failure caused by single color feature in complex scenarios,a dual color feature fusion target tracking method based on correlation filtering tracking algorithm is proposed.Firstly,the color histogram and low-dimensional adaptive color attributes(CN)of the target are extracted under the framework of correlated filtering target tracking,and the responses of different features are solved.Then,a new feature response fusion strategy is proposed.Then,a new feature response fusion strategy is proposed,which adapts the weight of feature response fusion through the distance relation of the target center of adjacent frames,reduces the adjustment of parameters in the response fusion phase,and increases the performance of the algorithm.(2)Aiming at the problem of poor robustness of the two-color algorithm in scenes such as illumination change,fast motion,background interference and so on.Based on the dual-color feature,the gradient histogram(HOG)feature is introduced,and a three-feature fusion tracking algorithm is proposed.The multi-feature fusion method enriches the feature representation of the target appearance and improves the tracking accuracy of the algorithm.In order to meet the real-time requirements of the tracking algorithm,the principal component analysis(PCA)technique is used to reduce the dimension of the extracted HOG features,which not only ensures the tracking accuracy but also reduces the computational complexity of the algorithm.(3)In order to solve the problem that the feature extraction of dense scale samples in the scale evaluation stage will increase the computational complexity,an improved scale evaluation strategy is adopted.Under the condition of ensuring the accuracy of target scale prediction,the processing speed of the algorithm is improved by reducing the number of scale samples and extracting the dimension of features.