Infrared Image Targets Detection And Tracking Algorithm Based On Compressed Sensing

In the field of image guidance,to improving the imaging resolution,the increasing of infrared focal plane array pixels number brings a series of problems for imaging equipment,such as the rising cost of manufacturing,uneven infrared response of pixels,and more complex reading circuit.The imaging equipments based on compressed sensing technology need only a few pixels which effectively reduces the processing difficulty of uniform pixel array.The traditional compressed sensing data processing method needs to reconstruct the original image before detecting and tracking targets,which takes up a large amount of computing resources and has an impact on the real-time performance of the system.In order to improve the detection speed,the method of target detection with no reconstructed is studied.For purpose of forming a complete target detection and tracking system based on compressed sensing,this paper further studies the tracking method for occluded targets.The main research work is as follows:Aiming at the poor real-time performance caused by completely reconstruct the original image,an infrared image targets detection and tracking framework is proposed.The framework divides the field of view with block compressed sensing and detects the target by using the statistical characteristics of the sampling results and the correlation characteristics between adjacent blocks.Then,based on the detection results,the framework gets the decompression data of target area by reconstructing blocks of target area and then tracks targets in the local region.It can not only make full use of image information to accurately track the location of the target,but also avoid a lot of calculation caused by complete images reconstruction.On this basis,the key parameters of block compressed sensing such as block size and compression rate are analyzed.Meanwhile,the guiding principles of parameter optimization are given.Simulation results show that the detection accuracy of the proposed framework is sensitive to block size.If the appropriate block size is selected,the detection accuracy is approximately the same as the traditional detection framework,but the detection speed is greatly improved.Compared with the target tracking method in complete reconstruction images,the computing speed of tracking in partial reconstruction images is significantly improved with the same tracking accuracy.Traditional infrared small target detection algorithms based on the assumed that targets' gray levels is higher than the local background region and can not adapt to the com-plex environment.In particular,it is impossible to make adaptive processing for bright and dark targets respectively.To solve this problem,this paper proposed a local segmentation contrast measure method based on brightness significance of images.In the proposed method,grey values of local pixels are divided into three segmentations.Different visual salience function are designed for dark targets and bright targets.On this basis,the accuracy and computational complexity of the proposed method are analyzed and the experiments are designed to verify the performance.The experiment results show that the proposed method can get a high detection rate on the premise of a low false alarm rate.Compared with the six typical preprocessing algorithms,the detection accuracy of local segmentation contrast measure algorithm is approximately equal to or even better than morphological filtering method.In terms of calculation speed,the proposed algorithm in this paper is much faster than the local contrast measure algorithm.The targets occluded by background occluder may lead to target template pollution in the online templates library updating.In order to solve this problem,an anti-occlusion target tracking strategy is proposed.The proposed robust spatio-temporal context(RSTC)method,inspired by the spatio-temporal context method,uses spatio-temporal context information to establish an early warning mechanism.Then it confirms the occlusion by detecting the change of the grey histogram of the target area and saves the target motion information and templates.Finally,the algorithm captures the target after the target reappears.Experimental results show that the abnormal detection mechanism is sensitive to occlusion and target maneuvering and can accurately confirm occlusion.In the relock stage,the proposed method can effectively use unpolluted information to capture the target.In conclusion the centre location errors of the proposed method are lower than the other nine tracking algorithms.The overlap success rates are higher than the other nine mainstream tracking algorithms.In the long-term heavy occlusion,as the characteristics of the target have changed,the algorithms failed to relock target,but correctly predicted the occlusion event.