The Study Of Multiple Target Detection And Tracking Based On Video Data Of Marine Radar System

Nowadays,radar plays an important role in modern national defence.The detection and tracking of targets in radar system are always hot topics.The resolution and the detection range of radars are always increasing as the deveolpment of radar systems.Some new problems and challenges have arise in the system for this reason.Meanwhile,the challenges are also opportunities to increas the performance of radar system.The input of the algorithms in this work is the video data of radar.The aim of this work is obtains the trakectories of the targets by the detetcion restlts of radar video data.The three topics which would been discussed here are given as follows.Firstly,with the increased resolution of moder radars,targets would be found in several resolution cells rather than merely appears in one single resolution cell,providing what is called an "extended targe"problem which including two parts:extended target detection problem and extended target tracking problem.Secondly,with the increased resolution and detection range of radars,the quantity of resolution cells increases significantly and the amount of calculation is huge.However,the raw data of one frame must be processed within a radar scanning cycle.Therefore the complexity and calculation of the method are significant.Lastly,as the increases of accurcy in extended target,much more information of targets and surveillance area can be obtained.The information is expected to improve the performance of detetcion and tarcking.The algorithms in image processing and bionicare are integrated into the problem of target detection and tracking in radar video data.The main content consists of the following parts:1.An algorithm to obtains the extended tragets efficiently is necessary.The increases of accuracy and detection range make the quantity of the data is enmorous.The speed of data processing is no larger than that of data acquirement with existing algorithms brings a tough problem in real time processing.The character of extended targets and the theory of?sample" are applied in the two algorthms which are peoposed in this work.An efficient extended target detection method based on contour trackingalgorithm and its improved version,an improved extended target detection method based on contour tracking and region growing algorithm are introduced.The extended targets in video data of radar can be well detected by the two methods.The proposed methodes and existing methods are tested with the real data which indicates that the efficiency of the two methodes are several times of those of existing methods.Meanwhile,the two proposed method can meet the acquirement of real time processing.2.The clutter is varous in different region.The surveillance area is very huge and complicated.Firstly,for the effect of the echoes of background(mountains,shores,buildings,islandsandmotor vehicles),enormous false alarm exists.Secondly,the average power of non-intentional interference varies in different areas.Therefore one single threshold is hard to obtain all the potential extended targets in various area.The extended targets in clutter region is misdetected by the two methods above.An improved target detection method which bases its principle on the multi contour tracking and region growing algorithm is proposed for the various conditions.Targets under different conditions can be detected by the multi contour tracking and region growing algorithm.Meanwhile,an efficient method based on spatiotemporal constant false-alarm rate(CFAR)thresholding and region growing algorithm is introduced.The detection thresholds of multi contour tracking and region growing algorithm are estimated by the parameters of radar and environment.However,the spatiotemporal CFAR method obtians the detection threshold by the previous measurements adaptively.Meanwhile,the adaptive capacity of the spatiotemporal CFAR method is better than that of the multi contour tracking and region growing algorithm under various conditions.Meanwhile,the spatiotemporal CFAR method is more easy to implement in parctice.3.The measurements of closely targets may be affected by the several targets simultaneously.For the width of beam and the measurement model,the measurement of a azimuth-range cell may be affected by several targets but the cell is not occupied by any targets.Meanwhile,the measurement of this cell is higher that those of other cells which are not occupied by any targets.Under this circumstances,the four methods above may regard the several tragets as one larger extended target.Thererofre,inspired by the theory of bionics,an algorithm providing what is called "rain algorithm" is proposed to improved the detection performance of closely targets.The "rain algorithm" can partition the measurements of targets into clusters.Each cluster denotes one extended target and the extended targets can be detected independently.4.The small and weak extended targets are hard to detect.The alogrithm introduced above only detect targets with one single frame.Plenty of misdetction and false alarm would arise when the targets are small and weak.Therefore,the theory of track before detect(TBD)is proposed to improve the detection performance.The targets in TBD algorithms are detected by several frame of measurements jointly.For the consideration of efficiency,the track before detect algorithm based on dynamic programming are utlized here rather than the particle filter based track before detect algorithm because the enormous particles bring plenty of calculation.A modified dynamic programming based track before detect(DP-TBD)which can track and detect extended targets are proposed.The proposed DP-TBD is also tested by the video data of a radar in real world,the results indicates that the extended targets in the video data of radars can be well detected by the DP-TBD efficiently.