Research On Infrared Dim Target Detection And Tracking Algorithm Under Deep-space Background

Infrared imaging technology is such a technology that passively receives the infrared radiation characteristics and thus acquisites the targets' images.It has a wide variety of advantages,including high concealment,full-time working,large observation range,and so on.With the increasing development and research of infrared imaging technology,infrared imaging systems have been widely used in diverse military or civil fields,such as monitoring,guidance,target tracking and searching,and so forth.Among these applications,detection and tracking of infrared dim targets is one of the most popular research hot spots.As we all know,it is a difficult task to detect and track the infrared dim targets under deep-space background due to its three unique features.Firstly,the observed target is a dim target and thus occupies only a few pixels in the image because of long imaging distance.Secondly,the target signature is usually weak and easily buried by the surrounding environment with high fluctuation under the interference of system noise and background noise.Thirdly,the target features that can be extracted from infrared image are significantly reduced since the targets in the image lack of fixed shape and texture information.Therefore,it is a challenging topic to detect and track the infrared dim targets under deep-space background,which accordingly has important theoretical and practical meanings.In this thesis,the related technologies of detection and tracking of infrared dim targets under deep-space background are studied.The main work and innovations are shown as follows:(1)Firstly,the infrared image preprocessing algorithm under deep-space background is first investigated deeply.An adaptive Butterworth high-pass filtering preprocessing algorithm based on variance-weighted information entropy(as an indicator of complexity)is proposed.The background complexity of deep-space infrared image is first assessed quantitatively based on the indicator of variance-weighted information entropy,and then the concept of background complexity of deep-space infrared image is suggested.Based on these,the coefficient of Butterworth high-pass filter is tuned by using the assessed parameter.Finally,the background images with different configurations,such as simple background,disturbed background with heavy noise and target-buried background,are processed adaptively.The experimental results show that the proposed algorithm can suppress the background noise well and hence improve the signal-to-noise ratio of the infrared image.(2)Secondly,the infrared dim target detection algorithm under deep-space background is studied in-depth.Aiming at the problem of poor real-time performance of traditional Otsu segmentation algorithm,a fast iterative algorithm for two-dimensional histogram oblique Otsu is proposed.In this algorithm,the advantages of two dimensional histogram oblique Otsu are analyzed comparing to two dimensional histogram vertical Otsu.And then,based on the idea of iteration and incorporating the target characteristics of infrared image,the algorithm is optimized by using energy accumulation.The extensive experimental results validate that the proposed algorithm can augment the target detection performance and meet the requirement of real-time performance simultaneously.(3)Thirdly,the infrared dim target tracking algorithm under deep-space background is also studied.An optimized auxiliary particle filter target tracking algorithm based on multi-feature fusion is presented.Aiming at the problems of shortage of target features and the difficulties in feature extraction,the effectiveness of the multi-feature fusion is first analyzed.And based on the idea of multi-feature fusion,the target feature information can be significantly augmented by combining different kinds of features.Then,the particle filter tracking algorithm which is originally designed for the nonlinear and non-Gaussian environment is adopted.In this algorithm,the weight updating scheme is optimized by adding an auxiliary variable,and the classical Mean-Shift algorithm is embedded in the filtering procedure.Finally,the overall target tracking procedure is optimized by combining the multi-feature fusion.The experimental results using simulated and real-world infrared images demonstrate that the proposed algorithm substantially improves the accuracy and robustness of infrared dim target tracking under deep-space background.