Research On Near-infrared Detection And Tracking Technology Of "low, Slow And Small" Targets In The Sky Background

The Open Air Policy is imminent for the past few years,and Low-Slow-Small(LLS)flight violations occur frequently,the early warning of LLS has become an important field in modern security construction.In this trend,a near-infrared detection platform is designed for LLS,the near-infrared image preprocessing and target tracking algorithm are studied and improved.(1)In view of the difficulties of photoelectric detection against sky background,nearinfrared imaging was chosen as the detection technology.In order to locate the target,the binocular localization scheme was studied and designed.The 360°PTZ is used to achieve the Omnibearing detection and tracking of the target,the hardware and software design of the PTZ tracking system was completed.(2)In view of the high dynamics of near-infrared imaging and the weak details of LLS against sky background,an image preprocessing algorithm based on bilateral filter layering was put forward.Based on the image data collected by the detection system in this paper,the nearinfrared imaging characteristics and target image features are analyzed.According to the analysis,the purpose of the image preprocessing algorithm is to compress the dynamic range and enhance the details of the image.Two classical dynamic range compression and enhancement algorithms,Automatic Gain Control(AGC)and Histogram Equalization(HE),are studied.Aiming at the defects of the two algorithms in detail protection,an image preprocessing algorithm based on bilateral filter layering is proposed.Experiments show that this algorithm can compress dynamic range and enhance detail,and it performs better than AGC and HE algorithms.(3)Aiming at the problem of occlusion and scale change when tracking the target,the Struck algorithm is studied and improved.We compared Struck,CSK and TLD algorithms on OTB50 dataset to verify the performance of Struck algorithm.We made four sets of target test sequences in different scenarios and verified the performance of Struck algorithm on the selfmade test sequences.According to the results,it is found that the Struck algorithm has poor robustness to scale change and lack the target state discrimination mechanism.In view of the above two defects,the adaptive scale mechanism and target state discrimination mechanism are introduced into the Struck algorithm.Experimental results show that the spatial robustness and accuracy of the modified algorithm are greatly improved,and the ability to identify the target state is obtained.(4)Based on the above research results,a test platform of tracking system was built,and the engineering implementation of image processing algorithm was optimized.On this test platform,we verified the feasibility of the tracking system in real world.