Research On Target Tracking Method Integrating Twin Network And Correlation Filtering

Target tracking provides reliable data for high-level applications,so it is widely used in intelligent medical,military,human-computer interaction and other aspects.In recent years,many researchers have conducted different levels of research on it.Although they have achieved relatively excellent research results,due to the influence of a variety of complex factors,the performance of target tracking is not very optimistic.Researching high-performance target tracking algorithms is still a challenging task.In order to better solve the problem that targets are prone to tracking drift under the influence of multiple complex factors,the paper first analyzes the current research status of target tracking algorithms;secondly,on this basis,design motion search strategies based on multiple model categories.And the construction of the observation model,focusing on the design of the model update strategy and the construction of the observation model;finally,the existing tracking model is improved,verified by a variety of test sequences,and the improved algorithm can produce better results after analysis.Tracking Results.The specific research work of this paper is as follows:1)Based on the background-sensing correlation filtering model,a background-sensing correlation filtering tracking algorithm with adaptive scale and learning rate adjustment is designed.First,introduce the scale-dependent filtering model,use the target's directional gradient histogram feature to effectively train the scale filter,use the scale-dependent filter to quantify the target scale during the tracking process,and obtain the scale change of the target;secondly,design the model framework When considering the interference of background information on the model,quantify the target scale change,so as to adjust the search area size proportionally;finally,introduce a high-confidence update strategy to determine the target occlusion based on the fluctuation of the response graph and the change of the maximum response value,Design the learning rate adjustment criteria and use the high-confidence update strategy to design the corresponding model update strategy to improve the generalization ability of the tracking model.2)Aiming at the problem that the fixed template cannot adapt to the appearance change of the target during the target movement,based on the framework of the fully convolutional twin neural network,a target tracking algorithm integrating the twin network and related filtering is designed.First,use the attention mechanism to process image information,use the spatial attention mechanism and the channel attention mechanism to optimize the target feature and the search area feature.In order to make full use of the context information of the search area image,the cross attention module is introduced to branch to the template.Then,in order to adapt to the change of the moving target,the relevant filtering model is introduced in the model construction process,which is regarded as a layer in the network model,so as to realize the update of the target template characteristics;finally,according to the response graph The confidence level selects the current location of the target to achieve accurate tracking of the target.3)The designed method is compared with other algorithms on the public tracking data set,and the tracking performance is analyzed.The experimental results show that the improved algorithm designed in this paper can effectively improve the tracking accuracy,and it also provides an improved reference for solving the problems often encountered in the target tracking algorithm.