Research On Detection And Tracking Of Dim And Small Target In Complex Infrared Image Sequences

With the rapid development of military science and technology,detection and tracking technology of infrared dim and small target plays an important role in infrared early warning,precision guidance,and long-range target detection.For long-distance infrared detection under the complex ground and cloud background,the dim and small target with texture-less information in the image is difficult to be distinguished.At the same time,strong edge information and image noise can easily cause interference to the target.Therefore,the infrared dim and small target detection and tracking algorithm must accurately detect the real target,eliminate interference to track the target robustly and achieve scene adaptability,which remains a challenging task.In response to these problems,we conduct in-depth research on the detection and tracking of the dim and small target in complex infrared image sequences.The main work and innovations of the paper are as follows:The infrared dim and small target image is analyzed from three aspects:the background,the noise,and dim and small target.The majority of the image are often occupied by the background,with its gradually changed gray scale which shows significant spatial correlation.The noise shows no relevance with the background and conforms to the Gaussian distribution on account of its random distribution in the space and non-correlated frames.The target is relatively small and texture-less,accounting for less than 0.12%of the entire image.And it presents as bright spots and follows an approximately Gaussian distribution.According to the characteristics of the background and the target,it can be considered that the background has low rank and the target has sparseness.An infrared dim and small target detection algorithm based on low rank and reweighted sparse representation is proposed.Firstly,the weights of local prior information are obtained using structure tensor,and the self-reinforced sparsity weights of the target matrix are extracted.Combined with the weight information,the sparsity of the target matrix is described by the weighted ₁L norm.Then the low-rank of the background matrix is described byγnorm and the noise matrix by L_2,1 norm Finally,the target is obtained through threshold segmentation and post-processing of the target matrix optimized by an ADMM algorithm over a constructed low-rank sparse representation model of the image.A large number of experiments have proved the superiority of the detection algorithm in this paper.An infrared dim and small target tracking algorithm based on motion features and depth appearance features is proposed.Firstly,the Kalman filter is used to extract the motion features of the dim and small target and to predict its position.Then,to alleviate the problem of mistracking and ID conversion,a Siamese convolutional network is designed to extract deep appearance features of target.Finally,a distance metric is constructed combining motion and deep appearance features,and the Hungarian matching algorithm is employed for data association.A large number of experiments show that the algorithm proposed in this paper has superiority.