Research On Tracking Algorithm With Correlation Scale Estimation Filtering And Multi-feature Fusion

Moving target tracking is an important direction of computer vision research.Due to the large amount of information in the video scene,the moving target contains most of the content of interest.In order to get more acciirate real-time tracking effect and more robust tracking algorithm,the research on moving target tracking algorithm has been progressing steadily.The problems of fast moving,out-of-plane rotation,illumination change,partial occlusion,etc.,which are widely used in video scenes,have always been the challenges that need to be overcome in the study of moving target tracking.Tracking research based on correlation filtering is a mainstream branch of the current moving target tracking field.The DSST(discriminative scale space tracking)algorithm proposed by Martin Danelljan et al.not only improves the tracking speed,but also has a good tracking effect.However,because the DSST algorithm uses the direction gradient histogram as a single target feature,the effect is not good when the target is deformed,fast motion and motion blur,and the background noise near the target is large,and the performance difference is different under different types of complex background sequences.The robustness is insufficient.In this thesis,from the perspective of reducing the interference of these problems in moving target in the actual application environment,based on the existing DSST algorithm,a DSST-based fusion feature improvement method is proposed,and the improvement effect of the method is verified by experiments.The main work done in this thesis is as follows:1?The general classification method of moving object tracking is introduced,and the principle and related knowledge of the technology used in this thesis are briefly explained,including directional gradient histogram,color name,color histogram and Gauss radial basis function.Then the classical correlation filtering algorithm MOSSE is briefly summarized,and then the algorithm idea of DSST tracking algorithm based on improved MOSSE is elaborated in detail..2.?Aiming at the shortcomings of DSST algorithm,this thesis adds two color features HSV and CN based on traditional DSST algorithm,and fuses HOG feature with three color features in dimensionality reduction.The fused feature is used as a new target feature for position estimation and scale estimation.An improved method based on DSST moving target tracking algorithm is proposed..3.?The method proposed in this thesis is analyzed and tested qualitatively and quantitatively by selecting typical international complex background test sets and relevant test standards.At the same time,DSST and ASMS algorithm are used as control group experiments.The experimental results show that the proposed method has better usability and robustness.