Research Of Moving Target Tracking Technology Based On Attention-LSTM Model

Target tracking is one of the most important research topics in the field of computer vision,and it has wide application and great development prospects in the fields of intelligent recognition,visual guidance and human-computer interaction.In order to solve the problem of target tracking,many researchers at home and abroad have made various efforts for many years and finally put forward many effective algorithms.However,how to track the target of apparent change in complex and changeable natural scenes is still a challenging issue.Common tracking difficulties include: scene lighting changes,the target moves,local cover,the target's status,the size and so on.In recent years,the convolution neural network theory has been successfully applied to the visual tracking algorithm,and has achieved good tracking effect,which has attracted the attention of the majority of researchers.focusing on these visual tracking of the difficulties,in this paper,we apply the visual attention mechanism and LSTM with circular convolution in the visual tracking algorithm to further improve the robustness,accuracy and real-time performance of the visual tracking algorithm.This paper firstly introduces the current research status of visual tracking technologies,then analyzes the application prospect of convolutional representation in the field of visual tracking.With the development of neural networks and big data,this paper presents a method which allocates different weights according to the role of each local image in expressing the target's apparent,and preprocesses the video sequence dataset,inputs it to the CNN for feature representation,then inputs it to the LSTM network structure for prediction.We incorporate the visual attention mechanism into the new network model that combines CNN with LSTM and use it for mobile target tracking.It applies the deep learning framework to extract the features of the video data,and uses double depth learning to synchronize space-flow convolution and time-flow convolution.At last,we use Matlab to simulate the Benchmark,the most authoritative moving target tracking algorithm test dataset,and compare the method mentioned in this paper with the current advanced tracking algorithms in many typical video sequences.Qualitative and quantitative experimental results show that the proposed algorithm can not only track the target accurately,but also has better overall performance than many advanced tracking algorithms in complex situations,such as occlusion,rotation,scale change,fast motion and illumination change.what's more,it achieves remarkable results in the tracking accuracy and success rate.