Research And Application Of Tracking Algorithm For Motion Target In Military Security

In recent years,with the development of high-tech,the process of building the informationization of the armed forces has also accelerated and all relevant units are stepping up their research and investment in military technology.In the field of military security,the research of motion target tracking technology has been a very extensive research field.Whether video surveillance systems,weapons and equipment research and development,or the construction of security defense systems,it is inseparable with support of the motion target tracking technology.With the rapid development of artificial intelligence and multimedia technology,the traditional motion target tracking technology has been unable to achieve high precision and high anti-interference performance.Therefore,based on the research and application of the motion target tracking algorithm in military security and defense,combined with popular deep learning technique and saliency target detection algorithm,this thesis proposes a novel salient motion target tracking algorithm based on deep learning framework.The new motion target tracking algorithm proposed in this thesis can not only have the characteristics of automatic learning of image features from deep learning,and avoid the deficiencies and one-sidedness of artificial design features and realize the accurate feature representation of the tracking target,but also introduce saliency detection technique into the motion target tracking technology to detect the saliency of the candidate image blocks and use the image block with the largest saliency area as the tracking target area of the next frame so as to successfully cope with the hidden and disguised targets in the tracking process.The main contribution of this thesis is summarized as follows:(1)This work fully expounds the important role of video surveillance technology and motion target tracking technology in military security and military fields in theactual warfare,weaponry and security defense system construction.Based on the research on the previous research background of the subject,as well as the comprehensive research significance and practical application value,the research content of the subject is put forward,that is,the research and application of tracking algorithm for motion target in military security.Afterwards,the status quo of the research at home and abroad is systematically and comprehensively analyzed and summarized.At the same time,we classified the motion target tracking algorithms which are popular in these years.In addition,the interference factors often encountered in the research of the motion target tracking algorithm are mainly introduced in six aspects.The mainstream motion target tracking databases are used to introduce these usual types of interference in the tracking process,which makes the research difficulty have a clearer understanding and more accurate positioning.(2)In this thesis,as the most important part,target feature representation module for motion target tracking task is analyzed.From two perspectives,we elaborate and analyze respectively,that is,the traditional image features and depth image features,and summarize the feature representation applied to the motion target tracking algorithm.In the traditional representation and extraction of image features,this study gives a brief introduction of the feature extraction methods for the popular color histogram features,LBP features,Haar features and SIFT features.And for the recent years,the hot deep image features are also introduced and analyzed.The design process of deep image feature is introduced and compared with the currently popular stacked denoising autoencoder(SDAE)and convolutional neural network(CNN).(3)In order to complete the task of researching and applying the motion target tracking algorithm in the military security,this work presents a clear and comprehensive introduction to the whole design process of the salient motion target tracking algorithm based on the deep learning framework.First,a convolutional neural network model is proposed,and a detailed model hierarchy and detail introduction are given.In addition,it explains how to use the large-scale labelled image database Imagenet to train the deep CNN model,and introduces the division ofrelevant data sets and the setting of training parameters.Then,in the process of representation and calculation of salient motion target,this paper introduces how to use the saliency detection algorithm to perfect and realize the detection of motion target,and shows the significant effect through examples.It can be found that the saliency detection algorithm can well realize the discrimination of the foreground of the tracking target in the moving target tracking algorithm and can well prevent the phenomenon of missing the target.Finally,the overall design flow of the motion target tracking algorithm is given,including how to track the motion target until the end of the last frame.(4)The proposed CNN-based salient motion target tracking algorithm is tested and applied in military security.This study compares all of the 29 tracking algorithms mentioned in object tracking Benchmark,with three major object tracking databases of OTB,VOT,and Temple Color.And,for the security work of military,this thesis selects the application scene similar to the security of the army from many open video tracking sequences and tests the application scenario of the proposed algorithm from four aspects.In addition to the tracking accuracy of the test results,the effect of saliency detection and labeling is very accurate,which is enough to illustrate the successful testing of the fusion with deep learning framework and saliency detection algorithm for motion target tracking.Finally,from a quantitative point of view,we compared the performance of all 30 algorithms in tracking accuracy and tracking overlap rate over multiple representative OTB tracking video sequences,and verified that the proposed motion target tracking algorithm can achieve very good tracking performance.