Research On Infrared Small Multi-Target Tracking Based On Spatio-temporal Correlation

Infrared dim small multi-target tracking has great significance in the military field.The infrared dim small multi-target tracking has the following problems under the sky background in the range,Firstly,the image signal-to-noise ratio is low,and the target size is small,then the available information is less,the background is complicated,and the tracking problem becomes difficult.Secondly,the phenomenon of trajectory cross-separation and target merging occurs in the cross movement is also one of the difficulties of multi-target tracking.Therefore,how to solve the influence of cloud background and achieve multi-target robust tracking is an urgent problem needs to solve.An improved bilateral filtering algorithm is proposed.For the interference of high frequency background information on the target extraction in the image,this dissertation solves the problem of background suppression in cloud background based on band-pass morphological filtering and windowing bilateral filtering,which greatly reduces the influence of high-frequency edge of cloud.Then the target is extracted via threshold segmentation by adaptive threshold method.A single-objective tracking method of spatio-temporal correlation based on the combination of weighted guided filtering and PSR decision is proposed.In the infrared target tracking,we analyze the drift situation caused by the algorithm of STC,and there are two main points,the first one is the low dimension of the prior knowledge,and the information on the gray level is simply collected and the high frequency information of background is not filtered out,without processing,the wrong information will be updated to the target model,which may lead to the error response of the confidence map,that is,tracking failed,the second is the fixed rate of learning,which leads to low adaptability of tracking.Based on this,the weighted guided image filtering and PSR discrimination mechanism are proposed to process the single target tracking based on spatio-temporal correlation.The former is used for filtering high-frequency background to provide optimized priori knowledge,the latter is used to determine the occlusion state and dynamically update the learning rate,and the effectiveness of the algorithm is verified by simulation experiment.The data association problem under multi-target tracking is also studied.To solve the problem of trajectory cross-separation and target merging between targets,a method of data association based on the continuous feature of the target is proposed,and an adaptive multi-target tracking framework is proposed in combination with the method of single target tracking based on spatio-temporal correlation.Under this framework,the tracking mode can be selected adaptively,which increases the robustness of the system.