Research On Detection And Tracking Technology Of Infrared Multi-target

The detection and tracking of weak targets on infrared video sequences has great significance in military fields such as investigation,surveillance,precision guidance and civilian fields such as civil aviation and meteorology.Due to the imaging characteristics of infrared images,generally the images quality is poor,the signal-to-clutter ratio and contrast ratio is low,the size of target area on infrared image is small,lacking features like shape and texture simultaneously,so compared with other multitarget tracking tasks,the detection and tracking of small targets on infrared images is more difficult.However,the existing multi-target tracking algorithms are hard to apply to this problem.In view of the difficulties in tracking small infrared targets listed above,this thesis proposes a complete algorithm step of detection before tracking.Firstly,in the multi-target detection step of infrared image,this thesis proposes a target detection algorithm based on an anisotropic spatial-temporal fourth-order diffusion filter.The proposed filter method utilizes the information of image sequences in time domain,realizes anisotropic filtering of image on edge direction and gray direction respectively,and obtain the target map through the filtered prediction background.By simulating multi-target detection tasks on multiple infrared backgrounds,it is verified that the proposed algorithm can obtain processed images with higher background suppression factor(BSF)and signal-to-clutter ratio gain(SCRG)than other state-of-the-art methods based on filters,which can prove that it has good performance.Secondly,under the premise of multiple target on each frame detected,this thesis utilizes a multitarget tracking method based on hypergraph matching,which solves the problem that it is difficult to establish mathematical model perfectly because small targets on infrared images have no obvious shape and texture features.In the light of the problem that the discrete property of existing hypergraph matching model,which result on the difficulty of achieving optimal solution in the process of continuous relaxation,a model with entropy barrier function term is proposed.Moreover,the model is solved by the improved non-monotone active set projected Newton method,which can get a more flexible stepsize during iterations and reduce the required time and space complexity.We calculated the matching accuracy of multi-target tracking on a simulation dataset frame by frame,and proved the effectiveness of our proposed model and improved optimization algorithm.In addition,this thesis also analyzes the special cases with false detection points,and experiments have verified that our algorithm has the ability to identify false detection targets.Finally,we make a summary of the work has been done.At the same time,the improvement of the target detection method based on an anisotropic spatial-temporal fourth-order diffusion filter is discussed,and the defects of multi-target tracking based on hypergraph matching are discussed,the work required in the future is forecasted.