Research And Implementation Of Infrared Single Target Tracking Method Based On Multiple Features

Visual target tracking is one of the important research fields of computer vision,and it also has important application value.Infrared imaging technology overcomes the imaging problem in weak illumination conditions,with the advantages of high imaging accuracy and ability of anti-interference.The application of infrared imaging technology in visual target tracking has greatly expanded the scope of visual target tracking.Infrared target tracking has been widely used in the fields of automated video surveillance,unmanned driving,military and weapon manufacturing.In recent years,with the continuous deepening of computer vision research,infrared single target tracking algorithm has made great progress.In practical applications,infrared single-target tracking still cannot achieve high precision due to factors such as background environment change,target deformation,or target occlusion.In this thesis,an infrared single target tracking method based on multi-features and correlation filter is proposed.According to the characteristics of infrared images,by dynamically combine various features,the impact of environmental changes is reduced,and the accuracy of target tracking is improved.Based on the tracking method,we designed and implemented an infrared single target tracking software.First,we introduced convolutional neural networks and the advantages of using convolutional features in correlated filter target tracking.We introduced the structure and characteristics of the VGG network,analyzed the characteristics of different depth convolution features.We selected a convolution feature suitable for tracking through experimental analysis.For the case that the convolutional features based correlation filter tracking method fails in some image sequences with similar target and background brightness,a method of using the motion information of the target to enhance the discrimination between the target and the background is proposed,the differential image is obtained by using the frame difference method,is used to extract the motion information of the target.The differential image based correlation filter tracking method can overcome the insufficiency of the convolution features to distinguish the target and the background in some sequences.Then,based on the advantages and disadvantages of the convolutional feature and the differential image feature,a method of dynamically combining the two features is proposed totake advantage of the two features.The convolutional feature and the differential image are used to train the correlation filter models.In the tracking phase,the features are input into the correlation filter models,and the obtained response maps are dynamically combined to obtain the final response map,and the final response map is used to locate the position of the target.Experiments show that the tracking accuracy of the tracking method combining the two features exceeds the tracking method using two features alone.We compare the method proposed in this thesis with several classical target tracking methods.Experiments show that the proposed method has higher accuracy than the comparative method in many sequences.Finally,according to the tracking method proposed in this thesis,we designed and implemented a infrared single target tracking software.By analyzing the functional requirements of the software,the functional module division of the software is determined,and the specific implementation of the software is introduced.In order to ensure the correctness and stability of the software,we have carried out software testing.The main functions of the software are:real-time visualization of tracking results,display of tracker running status,setting tracker parameters.